Search

Machine learning Filter

Boosting and AdaBoost in Machine Learning

Ensemble learning is a strategy in which a group of models are used to find a solution to a challenging problem, by using a strategy and combining diverse machine learning models into one single predictive model.In general, ensemble methods are mainly used for improving the overall performance accuracy of a model and combine several different models, also known as the base learners, to predict the results, instead of using a single model.In one of the articles related to ensemble learning, we have already discussed about the popular ensemble method, Bootstrap Aggregation. Bagging tries to implement similar learners on small sample populations and then takes a mean of all the predictions. It combines Bootstrapping and Aggregation to form one ensemble model. It basically reduces the variance error and helps to avoid overfitting. In this article we will look into the limitations of bagging and how a boosting algorithm can be used to overcome those limitations. We will also learn about various types of boosting algorithms and implement one of them in Python. Let’s get started.What are the limitations of Bagging?Let us recall the concept of bagging and consider a binary classification problem. We are either classifying an observation as 0 or as 1.In bagging, T bootstrap samples are selected, a classifier is fitted on each of these samples, and the models are trained in parallel. In a Random Forest, decision trees are trained in parallel. Then the results of all classifiers are averaged into a bagging classifier:Formula for a Bagging ClassifierLet us consider 3 classifiers and the result for the classification can either be right or wrong. If we plot the results of the 3 classifiers, there are regions in which the classifiers will be wrong. These regions are represented in red in the figure below.Example case in which Bagging works wellThe above example works pretty well as when one classifier is wrong, the two others are correct. By voting classifier, you can achieve a better accuracy. However, there are cases where Bagging does not work properly, when all classifiers are mistaken to be in the same region.Due to this reason, the intuition behind the discovery of Boosting was the following :instead of training parallel models, one should train models sequentiallyeach model should focus on where the performance of the previous classifier was poorWith this intuition, Boosting algorithm was introduced. Let us understand what Boosting is all about.What is Boosting?Boosting is an ensemble modeling technique which attempts to build a strong classifier from the number of weak classifiers. It is done by building a model using weak models in series. First, a model is built from the training data. Then the second model is built which tries to correct the errors present in the first model. This procedure is continued and models are added until either the complete training data set is predicted correctly or the maximum number of models are added.Boosting being a sequential process, each subsequent model attempts to correct the errors of the previous model. It is focused on reducing the bias unlike bagging. It makes the boosting algorithms prone to overfitting. To avoid overfitting, parameter tuning plays an important role in boosting algorithms, which will be discussed in the later part of this article. Some examples of boosting are XGBoost, GBM, ADABOOST etc..How can boosting identify weak learners?To find weak learners, we apply base learning (ML) algorithms with a different distribution. As each time base learning algorithm is applied, it generates a new weak prediction rule. This is an iterative process. After many iterations, the boosting algorithm combines these weak rules into a single strong prediction rule.How do we choose a different distribution for each round?Step 1: The base learner takes all the distributions and assigns equal weight or attention to each observation.Step 2: If there is any prediction error caused by first base learning algorithm, then we pay higher attention to observations having prediction error. Then, we apply the next base learning algorithm.Step 3: Iterate Step 2 till the limit of base learning algorithm is reached or higher accuracy is achieved.Finally, it combines the outputs from weak learner and creates a strong learner which eventually improves the prediction power of the model. Boosting gives higher focus to examples which are mis-classified or have higher errors by preceding weak rules.How would you classify an email as SPAM or not?Our initial approach would be to identify ‘SPAM’ and ‘NOT SPAM’ emails using the following criteria. If: Email has only one image file (promotional image), It’s a SPAM.Email has only link(s), It’s a SPAM.Email body consists of sentences like “You won a prize money of $ xxxxxx”, It’s a SPAM.Email from our official domain “www.knowledgehut.com” , Not a SPAM.Email from known source, Not a SPAM.Individually, these rules are not powerful enough to classify an email into ‘SPAM’ or ‘NOT SPAM’. Therefore, these rules are called as weak learner.To convert weak learner to strong learner, we’ll combine the prediction of each weak learner using methods like:Using average/ weighted averageConsidering prediction has higher voteExample: Above, we have defined 5 weak learners. Out of these 5, 3 are voted as ‘SPAM’ and 2 are voted as ‘Not a SPAM’. In this case, by default, we’ll consider an email as SPAM because we have higher(3) vote for ‘SPAM’Boosting helps in training a series of low performing algorithms, called weak learners, simply by adjusting the error metric over time. Weak learners are considered to be those algorithms whose error rate is slightly under 50% as illustrated below:Weighted errorsLet us consider data points on a 2D plot. Some of the data points will be well classified, others won’t. The weight attributed to each error when computing the error rate is 1/n where n is the number of data points to classify.Now if we apply some weight to the errors :You might now notice that we give more weight to the data points that are not well classified. An illustration of the weighting process is mentioned below:Example of weighting processIn the end, we want to build a strong classifier that may look like the figure mentioned below:Strong ClassifierTree stumpsThere might be a question in your mind about how many classifiers should one implement in order to ensure it works well. And how is each classifier chosen at each step?Well, Tree stumps defines a 1-level decision tree. At each step, we need to find the best stump, i.e the best data split, which will minimize the overall error. You can see a stump as a test, in which the assumption is that everything that lies on one side belongs to class 1, and everything that lies on the other side belongs to class 0.Many such combinations are possible for a tree stump. Let us look into an example to understand how many combinations we face.3 data points to splitWell there are 12 possible combinations. Let us check how.12 StumpsThere are 12 possible “tests” we could make. The “2” on the side of each separating line simply represents the fact that all points on one side could be points that belong to class 0, or to class 1. Therefore, there are 2 tests embedded in it.At each iteration t, we will choose ht the weak classifier that splits best the data, by reducing the overall error rate the most. Recall that the error rate is a modified error rate version that takes into account what has been introduced before.Finding the best splitThe best split is found by identifying at each iteration t, the best weak classifier ht, generally a decision tree with 1 node and 2 leaves (a stump). Let us consider an example of credit defaulter, i.e whether a person who borrowed money will return or not.Identifying the best splitIn this case, the best split at time t is to stump on the Payment history, since the weighted error resulting from this split is minimum.Simply note that decision tree classifiers like these ones can in practice be deeper than a simple stump. This will be considered as a hyper-parameter.Combining classifiersIn the next step we combine the classifiers into a Sign classifier, and depending on which side of the frontier a point will stand, it is classified as 0 or 1. It can be achieved by:Combining classifiersYou can improve the classifier by adding weights on each classifier, to avoid giving the same importance to the different classifiers.AdaBoostPseudo-codePseudo-codeThe key elements to keep in mind are:Z is a constant whose role is to normalize the weights so that they add up to 1αt is a weight that we apply to each classifierThis algorithm is called AdaBoost or Adaptive Boosting. This is one of the most important algorithms among all boosting methods.ComputationBoosting algorithms are generally fast to train, although we consider every stump possible and compute exponentials recursively.Well, if we choose αt and Z properly, the weights that are supposed to change at each step simplify to:Weights after choice of α and ZTypes of Boosting AlgorithmsUnderlying engine used for boosting algorithms can be anything.  It can be decision stamp, margin-maximizing classification algorithm etc. There are many boosting algorithms which use other types of engines such as: AdaBoost (Adaptive Boosting)Gradient Tree BoostingXGBoostIn this article, we will focus on AdaBoost and Gradient Boosting followed by their respective Python codes and a little bit about XGBoost.Where are Boosted algorithms required?Boosted algorithms are mainly used when there is plenty of data to make a prediction and high predictive power is expected. It is used to reduce bias and variance in supervised learning. It combines multiple weak predictors to build strong predictor.The underlying engine used for boosting algorithms can be anything. For instance, AdaBoost is a boosting done on Decision stump. There are many other boosting algorithms which use other types of engine such as:GentleBoostGradient BoostingLPBoostBrownBoostAdaptive BoostingAdaptive Boosting, or most commonly known AdaBoost, is a Boosting algorithm. This algorithm uses the method to correct its predecessor. It pays more attention to under fitted training instances by the previous model. Thus, at every new predictor the focus is more on the complicated cases more than the others.It fits a sequence of weak learners on different weighted training data. It starts by predicting the original data set and gives equal weight to each observation. If prediction is incorrect using the first learner, then it gives higher weight to observation which have been predicted incorrectly. Being an iterative process, it continues to add learner(s) until a limit is reached in the number of models or accuracy.Mostly, AdaBoost uses decision stamps. But, we can use any machine learning algorithm as base learner if it accepts weight on training data set. We can use AdaBoost algorithms for both classification and regression problems.Let us consider the example of the image mentioned above. In order to build an AdaBoost classifier, consider that as a first base classifier a Decision Tree algorithm is trained to make predictions on our training data. Applying the following methodology of AdaBoost, the weight of the misclassified training instances is increased. Then the second classifier is trained and the updated weights are acknowledged. It repeats the procedure over and over again.At the end of every model prediction we end up boosting the weights of the misclassified instances so that the next model does a better job on them, and so on.This sequential learning technique might sound similar to Gradient Descent, except that instead of tweaking a single predictor’s parameter to minimize the cost function, AdaBoost adds predictors to the ensemble, gradually making it better.One disadvantage of this algorithm is that the model cannot be parallelized since each predictor can only be trained after the previous one has been trained and evaluated.Below are the steps for performing the AdaBoost algorithm:Initially, all observations are given equal weights.A model is built on a subset of data.Using this model, predictions are made on the whole dataset.Errors are calculated by comparing the predictions and actual values.While creating the next model, higher weights are given to the data points which were predicted incorrectly.Weights can be determined using the error value. For instance,the higher the error the more is the weight assigned to the observation.This process is repeated until the error function does not change, or the maximum limit of the number of estimators is reached.Hyperparametersbase_estimators: specify the base type estimator, i.e. the algorithm to be used as base learner.n_estimators: It defines the number of base estimators, where the default is 10 but you can increase it in order to obtain a better performance.learning_rate: same impact as in gradient descent algorithmmax_depth: Maximum depth of the individual estimatorn_jobs: indicates to the system how many processors it is allowed to use. Value of ‘-1’ means there is no limit;random_state: makes the model’s output replicable. It will always produce the same results when you give it a fixed value as well as the same parameters and training data.Now, let us take a quick look at how to use AdaBoost in Python using a simple example on handwritten digit recognition.import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.ensemble import AdaBoostClassifier from sklearn.tree import DecisionTreeClassifier from sklearn.metrics import accuracy_score from sklearn.model_selection import cross_val_score from sklearn.model_selection import cross_val_predict from sklearn.model_selection import train_test_split from sklearn.model_selection import learning_curve from sklearn.datasets import load_digitsLet us load the data :dataset = load_digits() X = dataset['data'] y = dataset['target']X contains arrays of length 64 which are simply flattened 8x8 images. The aim of this dataset is to recognize handwritten digits. Let’s take a look at a given handwritten digit:plt.imshow(X[4].reshape(8,8))If we stick to a Decision Tree Classifier of depth 1 (a stump), here’s how to implement AdaBoost classifier:reg_ada = AdaBoostClassifier(DecisionTreeClassifier(max_depth=1)) scores_ada = cross_val_score(reg_ada, X, y, cv=6) scores_ada.mean()0.2636257855582272And it should head a result of around 26%, which can largely be improved. One of the key parameters is the depth of the sequential decision tree classifiers. How does accuracy improve with depth of the decision trees?score = [] for depth in [1,2,10] : reg_ada = AdaBoostClassifier(DecisionTreeClassifier(max_depth=depth)) scores_ada = cross_val_score(reg_ada, X, y, cv=6) score.append(scores_ada.mean()) score[0.2636257855582272, 0.5902852679072207, 0.9527524912410157]And the maximal score is reached for a depth of 10 in this simple example, with an accuracy of 95.3%.Gradient BoostingThis is another very popular Boosting algorithm which works pretty similar to what we’ve seen for AdaBoost. Gradient Boosting works by sequentially adding the previous predictors underfitted predictions to the ensemble, ensuring the errors made previously are corrected.The difference lies in what it does with the underfitted values of its predecessor. Contrary to AdaBoost, which tweaks the instance weights at every interaction, this method tries to fit the new predictor to the residual errors made by the previous predictor.So that you can understand Gradient Boosting it is important to understand Gradient Descent first.Below are the steps for performing the Gradient Boosting algorithm:A model is built on a subset of data.Using this model, predictions are made on the whole dataset.Errors are calculated by comparing the predictions and actual values.A new model is created using the errors calculated as target variable. Our objective is to find the best split to minimize the error.The predictions made by this new model are combined with the predictions of the previous.New errors are calculated using this predicted value and actual value.This process is repeated until the error function does not change, or the maximum limit of the number of estimators is reached.Hyperparametersn_estimators: It controls the number of weak learners.Learning_rate: Controls the contribution of weak learners in the final combination. There is a trade-off between learning_rate and n_estimators.min_samples_split: Minimum number of observation which is required in a node to be considered for splitting. It is used to control overfitting.min_samples_leaf: Minimum samples required in a terminal or leaf node. Lower values should be chosen for imbalanced class problems since the regions in which the minority class will be in the majority will be very small.min_weight_fraction_leaf: similar to the previous but defines a fraction of the total number of observations instead of an integer.max_depth : maximum depth of a tree. Used to control overfitting.max_lead_nodes : maximum number of terminal leaves in a tree. If this is defined max_depth is ignored.max_features : number of features it should consider while searching for the best split.You can tune loss function for better performance.Implementation in PythonYou can find Gradient Boosting function in Scikit-Learn’s library.# for regression from sklearn.ensemble import GradientBoostingRegressor model = GradientBoostingRegressor(n_estimators=3,learning_rate=1) model.fit(X,Y) # for classification from sklearn.ensemble import GradientBoostingClassifier model = GradientBoostingClassifier() model.fit(X,Y)XGBoostXG Boost or Extreme Gradient Boosting is an advanced implementation of the Gradient Boosting. This algorithm has high predictive power and is ten times faster than any other gradient boosting techniques. Moreover, it includes a variety of regularization which reduces overfitting and improves overall performance.AdvantagesIt implements regularization which helps in reducing overfit (Gradient Boosting does not have);It implements parallel processing which is much faster than Gradient Boosting;Allows users to define custom optimization objectives and evaluation criteria adding a whole new dimension to the model;XGBoost has an in-built routine to handle missing values;XGBoost makes splits up to the max_depth specified and then starts pruning the tree backwards and removes splits beyond which there is no positive gain;XGBoost allows a user to run a cross-validation at each iteration of the boosting process and thus it is easy to get the exact optimum number of boosting iterations in a single run.Boosting algorithms represent a different machine learning perspective which is turning a weak model to a stronger one to fix its weaknesses. I hope this article helped you understand how boosting works.We have covered most of the topics related to algorithms in our series of machine learning blogs, click here. If you are inspired by the opportunities provided by machine learning, enroll in our  Data Science and Machine Learning Courses for more lucrative career options in this landscape.Your one-stop-shop for Machine Learning is just a click away. Access our live online training and find easy solutions to all your queries here.

Boosting and AdaBoost in Machine Learning

8756
Boosting and AdaBoost in Machine Learning

Ensemble learning is a strategy in which a group of models are used to find a solution to a challenging problem, by using a strategy and combining diverse machine learning models into one single predictive model.

In general, ensemble methods are mainly used for improving the overall performance accuracy of a model and combine several different models, also known as the base learners, to predict the results, instead of using a single model.

In one of the articles related to ensemble learning, we have already discussed about the popular ensemble method, Bootstrap Aggregation. Bagging tries to implement similar learners on small sample populations and then takes a mean of all the predictions. It combines Bootstrapping and Aggregation to form one ensemble model. It basically reduces the variance error and helps to avoid overfitting. In this article we will look into the limitations of bagging and how a boosting algorithm can be used to overcome those limitations. We will also learn about various types of boosting algorithms and implement one of them in Python. Let’s get started.

What are the limitations of Bagging?

Let us recall the concept of bagging and consider a binary classification problem. We are either classifying an observation as 0 or as 1.

In bagging, T bootstrap samples are selected, a classifier is fitted on each of these samples, and the models are trained in parallel. In a Random Forest, decision trees are trained in parallel. Then the results of all classifiers are averaged into a bagging classifier:

What are the limitations of BaggingFormula for a Bagging ClassifierLet us consider 3 classifiers and the result for the classification can either be right or wrong. If we plot the results of the 3 classifiers, there are regions in which the classifiers will be wrong. These regions are represented in red in the figure below.
Example case in which Bagging works wellExample case in which Bagging works wellThe above example works pretty well as when one classifier is wrong, the two others are correct. By voting classifier, you can achieve a better accuracy. However, there are cases where Bagging does not work properly, when all classifiers are mistaken to be in the same region.

Bagging limitationsDue to this reason, the intuition behind the discovery of Boosting was the following :

  • instead of training parallel models, one should train models sequentially
  • each model should focus on where the performance of the previous classifier was poor

With this intuition, Boosting algorithm was introduced. Let us understand what Boosting is all about.

What is Boosting?

Boosting is an ensemble modeling technique which attempts to build a strong classifier from the number of weak classifiers. It is done by building a model using weak models in series. First, a model is built from the training data. Then the second model is built which tries to correct the errors present in the first model. This procedure is continued and models are added until either the complete training data set is predicted correctly or the maximum number of models are added.

Boosting being a sequential process, each subsequent model attempts to correct the errors of the previous model. It is focused on reducing the bias unlike bagging. It makes the boosting algorithms prone to overfitting. To avoid overfitting, parameter tuning plays an important role in boosting algorithms, which will be discussed in the later part of this article. Some examples of boosting are XGBoost, GBM, ADABOOST etc..

How can boosting identify weak learners?

To find weak learners, we apply base learning (ML) algorithms with a different distribution. As each time base learning algorithm is applied, it generates a new weak prediction rule. This is an iterative process. After many iterations, the boosting algorithm combines these weak rules into a single strong prediction rule.

How do we choose a different distribution for each round?

Step 1: The base learner takes all the distributions and assigns equal weight or attention to each observation.
Step 2: If there is any prediction error caused by first base learning algorithm, then we pay higher attention to observations having prediction error. Then, we apply the next base learning algorithm.
Step 3: Iterate Step 2 till the limit of base learning algorithm is reached or higher accuracy is achieved.

Finally, it combines the outputs from weak learner and creates a strong learner which eventually improves the prediction power of the model. Boosting gives higher focus to examples which are mis-classified or have higher errors by preceding weak rules.

How would you classify an email as SPAM or not?

Our initial approach would be to identify ‘SPAM’ and ‘NOT SPAM’ emails using the following criteria. If: 

  1. Email has only one image file (promotional image), It’s a SPAM.
  2. Email has only link(s), It’s a SPAM.
  3. Email body consists of sentences like “You won a prize money of $ xxxxxx”, It’s a SPAM.
  4. Email from our official domain “www.knowledgehut.com” , Not a SPAM.
  5. Email from known source, Not a SPAM.

Individually, these rules are not powerful enough to classify an email into ‘SPAM’ or ‘NOT SPAM’. Therefore, these rules are called as weak learner.

To convert weak learner to strong learner, we’ll combine the prediction of each weak learner using methods like:

  • Using average/ weighted average
  • Considering prediction has higher vote

Example: Above, we have defined 5 weak learners. Out of these 5, 3 are voted as ‘SPAM’ and 2 are voted as ‘Not a SPAM’. In this case, by default, we’ll consider an email as SPAM because we have higher(3) vote for ‘SPAM’

Boosting helps in training a series of low performing algorithms, called weak learners, simply by adjusting the error metric over time. Weak learners are considered to be those algorithms whose error rate is slightly under 50% as illustrated below:

Classifier error rate

Weighted errors

Let us consider data points on a 2D plot. Some of the data points will be well classified, others won’t. The weight attributed to each error when computing the error rate is 1/n where n is the number of data points to classify.

Now if we apply some weight to the errors :

Weighted errors

You might now notice that we give more weight to the data points that are not well classified. An illustration of the weighting process is mentioned below:

Example of weighting process

Example of weighting process

In the end, we want to build a strong classifier that may look like the figure mentioned below:


Strong Classifier
Strong Classifier
Tree stumps

There might be a question in your mind about how many classifiers should one implement in order to ensure it works well. And how is each classifier chosen at each step?

Well, Tree stumps defines a 1-level decision tree. At each step, we need to find the best stump, i.e the best data split, which will minimize the overall error. You can see a stump as a test, in which the assumption is that everything that lies on one side belongs to class 1, and everything that lies on the other side belongs to class 0.

Many such combinations are possible for a tree stump. Let us look into an example to understand how many combinations we face.

Tree stumps

3 data points to split

Well there are 12 possible combinations. Let us check how.


Tree Stumps

12 Stumps

There are 12 possible “tests” we could make. The “2” on the side of each separating line simply represents the fact that all points on one side could be points that belong to class 0, or to class 1. Therefore, there are 2 tests embedded in it.


At each iteration t, we will choose ht the weak classifier that splits best the data, by reducing the overall error rate the most. Recall that the error rate is a modified error rate version that takes into account what has been introduced before.

Finding the best split

The best split is found by identifying at each iteration t, the best weak classifier ht, generally a decision tree with 1 node and 2 leaves (a stump). Let us consider an example of credit defaulter, i.e whether a person who borrowed money will return or not.

Finding the best split

Identifying the best split

In this case, the best split at time t is to stump on the Payment history, since the weighted error resulting from this split is minimum.


Simply note that decision tree classifiers like these ones can in practice be deeper than a simple stump. This will be considered as a hyper-parameter.

Combining classifiers

In the next step we combine the classifiers into a Sign classifier, and depending on which side of the frontier a point will stand, it is classified as 0 or 1. It can be achieved by:

Combining classifiers

Combining classifiers

You can improve the classifier by adding weights on each classifier, to avoid giving the same importance to the different classifiers.


AdaBoost

AdaBoost

Pseudo-codePseudo-code

Pseudo-code
The key elements to keep in mind are:


  • Z is a constant whose role is to normalize the weights so that they add up to 1
  • αt is a weight that we apply to each classifier

This algorithm is called AdaBoost or Adaptive Boosting. This is one of the most important algorithms among all boosting methods.

Computation

Boosting algorithms are generally fast to train, although we consider every stump possible and compute exponentials recursively.

Well, if we choose αt and Z properly, the weights that are supposed to change at each step simplify to:

Weights after choice of α and Z

Weights after choice of α and Z


Types of Boosting Algorithms

Underlying engine used for boosting algorithms can be anything.  It can be decision stamp, margin-maximizing classification algorithm etc. There are many boosting algorithms which use other types of engines such as: 

  1. AdaBoost (Adaptive Boosting)
  2. Gradient Tree Boosting
  3. XGBoost

In this article, we will focus on AdaBoost and Gradient Boosting followed by their respective Python codes and a little bit about XGBoost.

Where are Boosted algorithms required?

Boosted algorithms are mainly used when there is plenty of data to make a prediction and high predictive power is expected. It is used to reduce bias and variance in supervised learning. It combines multiple weak predictors to build strong predictor.

The underlying engine used for boosting algorithms can be anything. For instance, AdaBoost is a boosting done on Decision stump. There are many other boosting algorithms which use other types of engine such as:

  1. GentleBoost
  2. Gradient Boosting
  3. LPBoost
  4. BrownBoost

Adaptive Boosting

Adaptive Boosting, or most commonly known AdaBoost, is a Boosting algorithm. This algorithm uses the method to correct its predecessor. It pays more attention to under fitted training instances by the previous model. Thus, at every new predictor the focus is more on the complicated cases more than the others.

It fits a sequence of weak learners on different weighted training data. It starts by predicting the original data set and gives equal weight to each observation. If prediction is incorrect using the first learner, then it gives higher weight to observation which have been predicted incorrectly. Being an iterative process, it continues to add learner(s) until a limit is reached in the number of models or accuracy.

Mostly, AdaBoost uses decision stamps. But, we can use any machine learning algorithm as base learner if it accepts weight on training data set. We can use AdaBoost algorithms for both classification and regression problems.

Let us consider the example of the image mentioned above. In order to build an AdaBoost classifier, consider that as a first base classifier a Decision Tree algorithm is trained to make predictions on our training data. Applying the following methodology of AdaBoost, the weight of the misclassified training instances is increased. Then the second classifier is trained and the updated weights are acknowledged. It repeats the procedure over and over again.

At the end of every model prediction we end up boosting the weights of the misclassified instances so that the next model does a better job on them, and so on.

This sequential learning technique might sound similar to Gradient Descent, except that instead of tweaking a single predictor’s parameter to minimize the cost function, AdaBoost adds predictors to the ensemble, gradually making it better.

One disadvantage of this algorithm is that the model cannot be parallelized since each predictor can only be trained after the previous one has been trained and evaluated.

Below are the steps for performing the AdaBoost algorithm:

  1. Initially, all observations are given equal weights.
  2. A model is built on a subset of data.
  3. Using this model, predictions are made on the whole dataset.
  4. Errors are calculated by comparing the predictions and actual values.
  5. While creating the next model, higher weights are given to the data points which were predicted incorrectly.
  6. Weights can be determined using the error value. For instance,the higher the error the more is the weight assigned to the observation.
  7. This process is repeated until the error function does not change, or the maximum limit of the number of estimators is reached.

Hyperparameters

base_estimators: specify the base type estimator, i.e. the algorithm to be used as base learner.

n_estimators: It defines the number of base estimators, where the default is 10 but you can increase it in order to obtain a better performance.

learning_rate: same impact as in gradient descent algorithm

max_depth: Maximum depth of the individual estimator

n_jobsindicates to the system how many processors it is allowed to use. Value of ‘-1’ means there is no limit;

random_state: makes the model’s output replicable. It will always produce the same results when you give it a fixed value as well as the same parameters and training data.

Now, let us take a quick look at how to use AdaBoost in Python using a simple example on handwritten digit recognition.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import cross_val_predict
from sklearn.model_selection import train_test_split
from sklearn.model_selection import learning_curve
from sklearn.datasets import load_digits

Let us load the data :

dataset = load_digits()
X = dataset['data']
y = dataset['target']

X contains arrays of length 64 which are simply flattened 8x8 images. The aim of this dataset is to recognize handwritten digits. Let’s take a look at a given handwritten digit:

plt.imshow(X[4].reshape(8,8))

Hyperparameters

If we stick to a Decision Tree Classifier of depth 1 (a stump), here’s how to implement AdaBoost classifier:

reg_ada = AdaBoostClassifier(DecisionTreeClassifier(max_depth=1))
scores_ada = cross_val_score(reg_ada, X, y, cv=6)
scores_ada.mean()
0.2636257855582272

And it should head a result of around 26%, which can largely be improved. One of the key parameters is the depth of the sequential decision tree classifiers. How does accuracy improve with depth of the decision trees?

score = []
for depth in [1,2,10] :
reg_ada = AdaBoostClassifier(DecisionTreeClassifier(max_depth=depth))
scores_ada = cross_val_score(reg_ada, X, y, cv=6)
score.append(scores_ada.mean())
score
[0.2636257855582272, 0.5902852679072207, 0.9527524912410157]

And the maximal score is reached for a depth of 10 in this simple example, with an accuracy of 95.3%.

Gradient Boosting

This is another very popular Boosting algorithm which works pretty similar to what we’ve seen for AdaBoost. Gradient Boosting works by sequentially adding the previous predictors underfitted predictions to the ensemble, ensuring the errors made previously are corrected.

The difference lies in what it does with the underfitted values of its predecessor. Contrary to AdaBoost, which tweaks the instance weights at every interaction, this method tries to fit the new predictor to the residual errors made by the previous predictor.

So that you can understand Gradient Boosting it is important to understand Gradient Descent first.

Below are the steps for performing the Gradient Boosting algorithm:

  1. A model is built on a subset of data.
  2. Using this model, predictions are made on the whole dataset.
  3. Errors are calculated by comparing the predictions and actual values.
  4. A new model is created using the errors calculated as target variable. Our objective is to find the best split to minimize the error.
  5. The predictions made by this new model are combined with the predictions of the previous.
  6. New errors are calculated using this predicted value and actual value.
  7. This process is repeated until the error function does not change, or the maximum limit of the number of estimators is reached.

Hyperparameters

n_estimators: It controls the number of weak learners.
Learning_rate: Controls the contribution of weak learners in the final combination. There is a trade-off between learning_rate and n_estimators.
min_samples_split: Minimum number of observation which is required in a node to be considered for splitting. It is used to control overfitting.
min_samples_leaf: Minimum samples required in a terminal or leaf node. Lower values should be chosen for imbalanced class problems since the regions in which the minority class will be in the majority will be very small.
min_weight_fraction_leaf: similar to the previous but defines a fraction of the total number of observations instead of an integer.
max_depth : maximum depth of a tree. Used to control overfitting.
max_lead_nodes : maximum number of terminal leaves in a tree. If this is defined max_depth is ignored.
max_features : number of features it should consider while searching for the best split.

You can tune loss function for better performance.

Implementation in Python

You can find Gradient Boosting function in Scikit-Learn’s library.

# for regression
from sklearn.ensemble import GradientBoostingRegressor
model = GradientBoostingRegressor(n_estimators=3,learning_rate=1)
model.fit(X,Y)
# for classification
from sklearn.ensemble import GradientBoostingClassifier
model = GradientBoostingClassifier()
model.fit(X,Y)

XGBoost

XG Boost or Extreme Gradient Boosting is an advanced implementation of the Gradient Boosting. This algorithm has high predictive power and is ten times faster than any other gradient boosting techniques. Moreover, it includes a variety of regularization which reduces overfitting and improves overall performance.

Advantages

  • It implements regularization which helps in reducing overfit (Gradient Boosting does not have);
  • It implements parallel processing which is much faster than Gradient Boosting;
  • Allows users to define custom optimization objectives and evaluation criteria adding a whole new dimension to the model;
  • XGBoost has an in-built routine to handle missing values;
  • XGBoost makes splits up to the max_depth specified and then starts pruning the tree backwards and removes splits beyond which there is no positive gain;
  • XGBoost allows a user to run a cross-validation at each iteration of the boosting process and thus it is easy to get the exact optimum number of boosting iterations in a single run.

Boosting algorithms represent a different machine learning perspective which is turning a weak model to a stronger one to fix its weaknesses. I hope this article helped you understand how boosting works.

We have covered most of the topics related to algorithms in our series of machine learning blogs, click here. If you are inspired by the opportunities provided by machine learning, enroll in our  Data Science and Machine Learning Courses for more lucrative career options in this landscape.


Your one-stop-shop for Machine Learning is just a click away. Access our live online training and find easy solutions to all your queries here.

Priyankur

Priyankur Sarkar

Data Science Enthusiast

Priyankur Sarkar loves to play with data and get insightful results out of it, then turn those data insights and results in business growth. He is an electronics engineer with a versatile experience as an individual contributor and leading teams, and has actively worked towards building Machine Learning capabilities for organizations.

Join the Discussion

Your email address will not be published. Required fields are marked *

Suggested Blogs

Top Data Analytics Certifications

What is data analytics?In the world of IT, every small bit of data count; even information that looks like pure nonsense has its significance. So, how do we retrieve the significance from this data? This is where Data Science and analytics comes into the picture.  Data Analytics is a process where data is inspected, transformed and interpreted to discover some useful bits of information from all the noise and make decisions accordingly. It forms the entire basis of the social media industry and finds a lot of use in IT, finance, hospitality and even social sciences. The scope in data analytics is nearly endless since all facets of life deal with the storage, processing and interpretation of data.Why data analytics? Data Analytics in this Information Age has nearly endless opportunities since literally everything in this era hinges on the importance of proper processing and data analysis. The insights from any data are crucial for any business. The field of data Analytics has grown more than 50 times from the early 2000s to 2021. Companies specialising in banking, healthcare, fraud detection, e-commerce, telecommunication, infrastructure and risk management hire data analysts and professionals every year in huge numbers.Need for certification:Skills are the first and foremost criteria for a job, but these skills need to be validated and recognised by reputed organisations for them to impress a potential employer. In the field of Data Analytics, it is pretty crucial to show your certifications. Hence, an employer knows you have hands-on experience in the field and can handle the workload of a real-world setting beyond just theoretical knowledge. Once you get a base certification, you can work your way up to higher and higher positions and enjoy lucrative pay packages. Top Data Analytics Certifications Certified Analytics Professional (CAP) Microsoft Certified Azure Data Scientist Associate Cloudera Certified Associate (CCA) Data Analyst Associate Certified Analytics Professional (aCAP) SAS Certified Data Analyst (Using SAS91. Certified Analytics Professional (CAP)A certification from an organisation called INFORMS, CAP is a notoriously rigorous certification and stands out like a star on an applicant's resume. Those who complete this program gain an invaluable credential and are able to distinguish themselves from the competition. It gives a candidate a comprehensive understanding of the analytical process's various fine aspects--from framing hypotheses and analytic problems to the proper methodology, along with acquisition, model building and deployment process with long-term life cycle management. It needs to be renewed after three years.The application process is in itself quite complex, and it also involves signing the CAP Code of Ethics before one is given the certification. The CAP panel reviews each application, and those who pass this review are the only ones who can give the exam.  Prerequisite: A bachelor’s degree with 5 years of professional experience or a master's degree with 3 years of professional experience.  Exam Fee & Format: The base price is $695. For individuals who are members of INFORMS the price is $495. (Source) The pass percentage is 70%. The format is a four option MCQ paper. Salary: $76808 per year (Source) 2. Cloudera Certified Associate (CCA) Data Analyst Cloudera has a well-earned reputation in the IT sector, and its Associate Data analyst certification can help bolster the resume of Business intelligence specialists, system architects, data analysts, database administrators as well as developers. It has a specific focus on SQL developers who aim to show their proficiency on the platform.This certificate validates an applicant's ability to operate in a CDH environment by Cloudera using Impala and Hive tools. One doesn't need to turn to expensive tuitions and academies as Cloudera offers an Analyst Training course with almost the same objectives as the exam, leaving one with a good grasp of the fundamentals.   Prerequisites: basic knowledge of SQL and Linux Command line Exam Fee & Format: The cost of the exam is $295 (Source), The test is a performance-based test containing 8-12 questions to be completed in a proctored environment under 129 minutes.  Expected Salary: You can earn the job title of Cloudera Data Analyst that pays up to $113,286 per year. (Source)3. Associate Certified Analytics Professional (aCAP)aCAP is an entry-level certification for Analytics professionals with lesser experience but effective knowledge, which helps in real-life situations. It is for those candidates who have a master’s degree in a field related to data analytics.  It is one of the few vendor-neutral certifications on the list and must be converted to CAP within 6 years, so it offers a good opportunity for those with a long term path in a Data Analytics career. It also needs to be renewed every three years, like the CAP certification. Like its professional counterpart, aCAP helps a candidate step out in a vendor-neutral manner and drastically increases their professional credibility.  Prerequisite: Master’s degree in any discipline related to data Analytics. Exam Fee: The base price is $300. For individuals who are members of INFORMS the price is $200. (Source). There is an extensive syllabus which covers: i. Business Problem Framing, ii. Analytics Problem Framing, iii. Data, iv. Methodology Selection, v. Model Building, vi. Deployment, vii. Lifecycle Management of the Analytics process, problem-solving, data science and visualisation and much more.4. SAS Certified Data Analyst (Using SAS9)From one of the pioneers in IT and Statistics - the SAS Institute of Data Management - a SAS Certified Data Scientist can gain insights and analyse various aspects of data from businesses using tools like the SAS software and other open-source methodology. It also validates competency in using complex machine learning models and inferring results to interpret future business strategy and release models using the SAS environment. SAS Academy for Data Science is a viable institute for those who want to receive proper training for the exam and use this as a basis for their career.  Prerequisites: To earn this credential, one needs to pass 5 exams, two from the SAS Certified Big Data Professional credential and three exams from the SAS Certified Advanced Analytics Professional Credential. Exam Fee: The cost for each exam is $180. (Source) An exception is Predictive Modelling using the SAS Enterprise Miner, costing $250, This exam can be taken in the English language. One can join the SAS Academy for Data Science and also take a practice exam beforehand. Salary: You can get a job as a SAS Data Analyst that pays up to $90,000 per year! (Source) 5. IBM Data Science Professional CertificateWhenever someone studies the history of a computer, IBM (International Business Machines) is the first brand that comes up. IBM is still alive and kicking, now having forayed into and becoming a major player in the Big Data segment. The IBM Data Science Professional certificate is one of the beginner-level certificates if you want to sink your hands into the world of data analysis. It shows a candidate's skills in various topics pertaining to data sciences, including various open-source tools, Python databases, SWL, data visualisation, and data methodologies.  One needs to complete nine courses to earn the certificate. It takes around three months if one works twelve hours per week. It also involves the completion of various hands-on assignments and building a portfolio. A candidate earns the Professional certificate from Coursera and a badge from IBM that recognises a candidate's proficiency in the area. Prerequisites: It is the optimal course for freshers since it requires no requisite programming knowledge or proficiency in Analytics. Exam Fee: It costs $39 per month (Source) to access the course materials and the certificate. The course is handled by the Coursera organisation. Expected Salary: This certification can earn you the title of IBM Data Scientist and help you earn a salary of $134,846 per annum. (Source) 6. Microsoft Certified Azure Data Scientist AssociateIt's one of the most well-known certifications for newcomers to step into the field of Big Data and Data analytics. This credential is offered by the leader in the industry, Microsoft Azure. This credential validates a candidate's ability to work with Microsoft Azure developing environment and proficiency in analysing big data, preparing data for the modelling process, and then progressing to designing models. One advantage of this credential is that it has no expiry date and does not need renewal; it also authorises the candidate’s extensive knowledge in predictive Analytics. Prerequisites: knowledge and experience in data science and using Azure Machine Learning and Azure Databricks. Exam Fee: It costs $165 to (Source) register for the exam. One advantage is that there is no need to attend proxy institutions to prepare for this exam, as Microsoft offers free training materials as well as an instructor-led course that is paid. There is a comprehensive collection of resources available to a candidate. Expected Salary: The job title typically offered is Microsoft Data Scientist and it typically fetches a yearly pay of $130,993.(Source) Why be a Data Analytics professional? For those already working in the field of data, being a Data Analyst is one of the most viable options. The salary of a data analyst ranges from $65,000 to $85,000 depending on number of years of experience. This lucrative salary makes it worth the investment to get a certification and advance your skills to the next level so that you can work for multinational companies by interpreting and organising data and using this analysis to accelerate businesses. These certificates demonstrate that you have the required knowledge needed to operate data models of the volumes needed by big organizations. 1. Demand is more than supply With the advent of the Information Age, there has been a huge boom in companies that either entirely or partially deal with IT. For many companies IT forms the core of their business. Every business has to deal with data, and it is crucial to get accurate insights from this data and use it to further business interests and expand profits. The interpretation of data also aims to guide them in the future to make the best business decisions.  Complex business intelligence algorithms are in place these days. They need trained professionals to operate them; since this field is relatively new, there is a shortage of experts. Thus, there are vacancies for data analyst positions with lucrative pay if one is qualified enough.2. Good pay with benefitsA data analyst is an extremely lucrative profession, with an average base pay of $71,909 (Source), employee benefits, a good work-home balance, and other perks. It has been consistently rated as being among the hottest careers of the decade and allows professionals to have a long and satisfying career.   Companies Hiring Certified Data Analytics Professionals Oracle A California based brand, Oracle is a software company that is most famous for its data solutions. With over 130000 employees and a revenue of 39 billion, it is surely one of the bigger players in Data Analytics.  MicroStrategy   Unlike its name, this company is anything but micro, with more than 400 million worth of revenue. It provides a suite of analytical products along with business mobility solutions. It is a key player in the mobile space, working natively with Android and iOS.   SAS   One of the companies in the list which provides certifications and is also without a doubt one of the largest names in the field of Big Data, machine learning and Data Analytics, is SAS. The name SAS is derived from Statistical Analysis System. This company is trusted and has a solid reputation. It is also behind the SAS Institute for Data Science. Hence, SAS is the organisation you would want to go to if you're aiming for a long-term career in data science.    Conclusion To conclude, big data and data Analytics are a field of endless opportunities. By investing in the right credential, one can pave the way to a viable and lucrative career path. Beware though, there are lots of companies that provide certifications, but only recognised and reputed credentials will give you the opportunities you are seeking. Hiring companies look for these certifications as a mark of authenticity of your hands-on experience and the amount of work you can handle effectively. Therefore, the credential you choose for yourself plays a vital role in the career you can have in the field of Data analytics.  Happy learning!    
5631
Top Data Analytics Certifications

What is data analytics?In the world of IT, every s... Read More

Why Should You Start a Career in Machine Learning?

If you are even remotely interested in technology you would have heard of machine learning. In fact machine learning is now a buzzword and there are dozens of articles and research papers dedicated to it.  Machine learning is a technique which makes the machine learn from past experiences. Complex domain problems can be resolved quickly and efficiently using Machine Learning techniques.  We are living in an age where huge amounts of data are produced every second. This explosion of data has led to creation of machine learning models which can be used to analyse data and to benefit businesses.  This article tries to answer a few important concepts related to Machine Learning and informs you about the career path in this prestigious and important domain.What is Machine Learning?So, here’s your introduction to Machine Learning. This term was coined in the year 1997. “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at the tasks improves with the experiences.”, as defined in the book on ML written by Mitchell in 1997. The difference between a traditional programming and programming using Machine Learning is depicted here, the first Approach (a) is a traditional approach, and second approach (b) is a Machine Learning based approach.Machine Learning encompasses the techniques in AI which allow the system to learn automatically looking at the data available. While learning, the system tries to improve the experience without making any explicit efforts in programming. Any machine learning application follows the following steps broadlySelecting the training datasetAs the definition indicates, machine learning algorithms require past experience, that is data, for learning. So, selection of appropriate data is the key for any machine learning application.Preparing the dataset by preprocessing the dataOnce the decision about the data is made, it needs to be prepared for use. Machine learning algorithms are very susceptible to the small changes in data. To get the right insights, data must be preprocessed which includes data cleaning and data transformation.  Exploring the basic statistics and properties of dataTo understand what the data wishes to convey, the data engineer or Machine Learning engineer needs to understand the properties of data in detail. These details are understood by studying the statistical properties of data. Visualization is an important process to understand the data in detail.Selecting the appropriate algorithm to apply on the datasetOnce the data is ready and understood in detail, then appropriate Machine Learning algorithms or models are selected. The choice of algorithm depends on characteristics of data as well as type of task to be performed on the data. The choice also depends on what kind of output is required from the data.Checking the performance and fine-tuning the parameters of the algorithmThe model or algorithm chosen is fine-tuned to get improved performance. If multiple models are applied, then they are weighed against the performance. The final algorithm is again fine-tuned to get appropriate output and performance.Why Pursue a Career in Machine Learning in 2021?A recent survey has estimated that the jobs in AI and ML have grown by more than 300%. Even before the pandemic struck, Machine Learning skills were in high demand and the demand is expected to increase two-fold in the near future.A career in machine learning gives you the opportunity to make significant contributions in AI, the future of technology. All the big and small businesses are adopting Machine Learning models to improve their bottom-line margins and return on investment.  The use of Machine Learning has gone beyond just technology and it is now used in diverse industries including healthcare, automobile, manufacturing, government and more. This has greatly enhanced the value of Machine Learning experts who can earn an average salary of $112,000.  Huge numbers of jobs are expected to be created in the coming years.  Here are a few reasons why one should pursue a career in Machine Learning:The global machine learning market is expected to touch $20.83B in 2024, according to Forbes.  We are living in a digital age and this explosion of data has made the use of machine learning models a necessity. Machine Learning is the only way to extract meaning out of data and businesses need Machine Learning engineers to analyze huge data and gain insights from them to improve their businesses.If you like numbers, if you like research, if you like to read and test and if you have a passion to analyse, then machine learning is the career for you. Learning the right tools and programming languages will help you use machine learning to provide appropriate solutions to complex problems, overcome challenges and grow the business.Machine Learning is a great career option for those interested in computer science and mathematics. They can come up with new Machine Learning algorithms and techniques to cater to the needs of various business domains.As explained above, a career in machine learning is both rewarding and lucrative. There are huge number of opportunities available if you have the right expertise and knowledge. On an average, Machine Learning engineers get higher salaries, than other software developers.Years of experience in the Machine Learning domain, helps you break into data scientist roles, which is not just among the hottest careers of our generation but also a highly respected and lucrative career. Right skills in the right business domain helps you progress and make a mark for yourself in your organization. For example, if you have expertise in pharmaceutical industries and experience working in Machine learning, then you may land job roles as a data scientist consultant in big pharmaceutical companies.Statistics on Machine learning growth and the industries that use MLAccording to a research paper in AI Multiple (https://research.aimultiple.com/ml-stats/), the Machine Learning market will grow to 9 Billion USD by the end of 2022. There are various areas where Machine Learning models and solutions are getting deployed, and businesses see an overall increase of 44% investments in this area. North America is one of the leading regions in the adoption of Machine Learning followed by Asia.The Global Machine Learning market will grow by 42% which is evident from the following graph. Image sourceThere is a huge demand for Machine Learning modelling because of the large use of Cloud Based Applications and Services. The pandemic has changed the face of businesses, making them heavily dependent on Cloud and AI based services. Google, IBM, and Amazon are just some of the companies that have invested heavily in AI and Machine Learning based application development, to provide robust solutions for problems faced by small to large scale businesses. Machine Learning and Cloud based solutions are scalable and secure for all types of business.ML analyses and interprets data patterns, computing and developing algorithms for various business purposes.Advantages of Machine Learning courseNow that we have established the advantages of perusing a career in Machine Learning, let’s understand from where to start our machine learning journey. The best option would be to start with a Machine Learning course. There are various platforms which offer popular Machine Learning courses. One can always start with an online course which is both effective and safe in these COVID times.These courses start with an introduction to Machine Learning and then slowly help you to build your skills in the domain. Many courses even start with the basics of programming languages such as Python, which are important for building Machine Learning models. Courses from reputed institutions will hand hold you through the basics. Once the basics are clear, you may switch to an offline course and get the required certification.Online certifications have the same value as offline classes. They are a great way to clear your doubts and get personalized help to grow your knowledge. These courses can be completed along with your normal job or education, as most are self-paced and can be taken at a time of your convenience. There are plenty of online blogs and articles to aid you in completion of your certification.Machine Learning courses include many real time case studies which help you in understanding the basics and application aspects. Learning and applying are both important and are covered in good Machine Learning Courses. So, do your research and pick an online tutorial that is from a reputable institute.What Does the Career Path in Machine Learning Look Like?One can start their career in Machine Learning domain as a developer or application programmer. But the acquisition of the right skills and experience can lead you to various career paths. Following are some of the career options in Machine Learning (not an exhaustive list):Data ScientistA data scientist is a person with rich experience in a particular business field. A person who has a knowledge of domain, as well as machine learning modelling, is a data scientist. Data Scientists’ job is to study the data carefully and suggest accurate models to improve the business.AI and Machine Learning EngineerAn AI engineer is responsible for choosing the proper Machine Learning Algorithm based on natural language processing and neural network. They are responsible for applying it in AI applications like personalized advertising.  A Machine Learning Engineer is responsible for creating the appropriate models for improvement of the businessData EngineerA Data Engineer, as the name suggests, is responsible to collect data and make it ready for the application of Machine Learning models. Identification of the right data and making it ready for extraction of further insights is the main work of a data engineer.Business AnalystA person who studies the business and analyzes the data to get insights from it is a Business Analyst. He or she is responsible for extracting the insights from the data at hand.Business Intelligence (BI) DeveloperA BI developer uses Machine Learning and Data Analytics techniques to work on a large amount of data. Proper representation of data to suit business decisions, using the latest tools for creation of intuitive dashboards is the role of a BI developer.  Human Machine Interface learning engineerCreating tools using machine learning techniques to ease the human machine interaction or automate decisions, is the role of a Human Machine Interface learning engineer. This person helps in generating choices for users to ease their work.Natural Language Processing (NLP) engineer or developerAs the name suggests, this person develops various techniques to process Natural Language constructs. Building applications or systems using machine learning techniques to build Natural Language based applications is their main task. They create multilingual Chatbots for use in websites and other applications.Why are Machine Learning Roles so popular?As mentioned above, the market growth of AI and ML has increased tremendously over the past years. The Machine Learning Techniques are applied in every domain including marketing, sales, product recommendations, brand retention, creating advertising, understanding the sentiments of customer, security, banking and more. Machine learning algorithms are also used in emails to ease the users work. This says a lot, and proves that a career in Machine Learning is in high demand as all businesses are incorporating various machine learning techniques and are improving their business.One can harness this popularity by skilling up with Machine Learning skills. Machine Learning models are now being used by every company, irrespective of their size--small or big, to get insights on their data and use these insights to improve the business. As every company wishes to grow faster, they are deploying more machine learning engineers to get their work done on time. Also, the migration of businesses to Cloud services for better security and scalability, has increased their requirement for more Machine Learning algorithms and models to cater to their needs.Introducing the Machine learning techniques and solutions has brought huge returns for businesses.  Machine Learning solution providers like Google, IBM, Microsoft etc. are investing in human resources for development of Machine Learning models and algorithms. The tools developed by them are popularly used by businesses to get early returns. It has been observed that there is significant increase in patents in Machine Learning domains since the past few years, indicating the quantum of work happening in this domain.Machine Learning SkillsLet’s visit a few important skills one must acquire to work in the domain of Machine Learning.Programming languagesKnowledge of programming is very important for a career in Machine Learning. Languages like Python and R are popularly used to develop applications using Machine Learning models and algorithms. Python, being the simplest and most flexible language, is very popular for AI and Machine Learning applications. These languages provide rich support of libraries for implementation of Machine Learning Algorithms. A person who is good in programming can work very efficiently in this domain.Mathematics and StatisticsThe base for Machine Learning is mathematics and statistics. Statistics applied to data help in understanding it in micro detail. Many machine learning models are based on the probability theory and require knowledge of linear algebra, transformations etc. A good understanding of statistics and probability increases the early adoption to Machine Learning domain.Analytical toolsA plethora of analytical tools are available where machine learning models are already implemented and made available for use. Also, these tools are very good for visualization purposes. Tools like IBM Cognos, PowerBI, Tableue etc are important to pursue a career as a  Machine Learning engineer.Machine Learning Algorithms and librariesTo become a master in this domain, one must master the libraries which are provided with various programming languages. The basic understanding of how machine learning algorithms work and are implemented is crucial.Data Modelling for Machine Learning based systemsData lies at the core of any Machine Learning application. So, modelling the data to suit the application of Machine Learning algorithms is an important task. Data modelling experts are the heart of development teams that develop machine learning based systems. SQL based solutions like Oracle, SQL Server, and NoSQL solutions are important for modelling data required for Machine Learning applications. MongoDB, DynamoDB, Riak are some important NOSQL based solutions available to process unstructured data for Machine Learning applications.Other than these skills, there are two other skills that may prove to be beneficial for those planning on a career in the Machine Learning domain:Natural Language processing techniquesFor E-commerce sites, customer feedback is very important and crucial in determining the roadmap of future products. Many customers give reviews for the products that they have used or give suggestions for improvement. These feedbacks and opinions are analyzed to gain more insights about the customers buying habits as well as about the products. This is part of natural language processing using Machine Learning. The likes of Google, Facebook, Twitter are developing machine learning algorithms for Natural Language Processing and are constantly working on improving their solutions. Knowledge of basics of Natural Language Processing techniques and libraries is must in the domain of Machine Learning.Image ProcessingKnowledge of Image and Video processing is very crucial when a solution is required to be developed in the area of security, weather forecasting, crop prediction etc. Machine Learning based solutions are very effective in these domains. Tools like Matlab, Octave, OpenCV are some important tools available to develop Machine Learning based solutions which require image or video processing.ConclusionMachine Learning is a technique to automate the tasks based on past experiences. This is among the most lucrative career choices right now and will continue to remain so in the future. Job opportunities are increasing day by day in this domain. Acquiring the right skills by opting for a proper Machine Learning course is important to grow in this domain. You can have an impressive career trajectory as a machine learning expert, provided you have the right skills and expertise.
5684
Why Should You Start a Career in Machine Learning?

If you are even remotely interested in technology ... Read More

Types of Probability Distributions Every Data Science Expert Should know

Data Science has become one of the most popular interdisciplinary fields. It uses scientific approaches, methods, algorithms, and operations to obtain facts and insights from unstructured, semi-structured, and structured datasets. Organizations use these collected facts and insights for efficient production, business growth, and to predict user requirements. Probability distribution plays a significant role in performing data analysis equipping a dataset for training a model. In this article, you will learn about the types of Probability Distribution, random variables, types of discrete distributions, and continuous distribution.  What is Probability Distribution? A Probability Distribution is a statistical method that determines all the probable values and possibilities that a random variable can deliver from a particular range. This range of values will have a lower bound and an upper bound, which we call the minimum and the maximum possible values.  Various factors on which plotting of a value depends are standard deviation, mean (or average), skewness, and kurtosis. All of these play a significant role in Data science as well. We can use probability distribution in physics, engineering, finance, data analysis, machine learning, etc. Significance of Probability distributions in Data Science In a way, most of the data science and machine learning operations are dependent on several assumptions about the probability of your data. Probability distribution allows a skilled data analyst to recognize and comprehend patterns from large data sets; that is, otherwise, entirely random variables and values. Thus, it makes probability distribution a toolkit based on which we can summarize a large data set. The density function and distribution techniques can also help in plotting data, thus supporting data analysts to visualize data and extract meaning. General Properties of Probability Distributions Probability distribution determines the likelihood of any outcome. The mathematical expression takes a specific value of x and shows the possibility of a random variable with p(x). Some general properties of the probability distribution are – The total of all probabilities for any possible value becomes equal to 1. In a probability distribution, the possibility of finding any specific value or a range of values must lie between 0 and 1. Probability distributions tell us the dispersal of the values from the random variable. Consequently, the type of variable also helps determine the type of probability distribution.Common Data Types Before jumping directly into explaining the different probability distributions, let us first understand the different types of probability distributions or the main categories of the probability distribution. Data analysts and data engineers have to deal with a broad spectrum of data, such as text, numerical, image, audio, voice, and many more. Each of these have a specific means to be represented and analyzed. Data in a probability distribution can either be discrete or continuous. Numerical data especially takes one of the two forms. Discrete data: They take specific values where the outcome of the data remains fixed. Like, for example, the consequence of rolling two dice or the number of overs in a T-20 match. In the first case, the result lies between 2 and 12. In the second case, the event will be less than 20. Different types of discrete distributions that use discrete data are: Binomial Distribution Hypergeometric Distribution Geometric Distribution Poisson Distribution Negative Binomial Distribution Multinomial Distribution  Continuous data: It can obtain any value irrespective of bound or limit. Example: weight, height, any trigonometric value, age, etc. Different types of continuous distributions that use continuous data are: Beta distribution Cauchy distribution Exponential distribution Gamma distribution Logistic distribution Weibull distribution Types of Probability Distribution explained Here are some of the popular types of Probability distributions used by data science professionals. (Try all the code using Jupyter Notebook) Normal Distribution: It is also known as Gaussian distribution. It is one of the simplest types of continuous distribution. This probability distribution is symmetrical around its mean value. It also shows that data at close proximity of the mean is frequently occurring, compared to data that is away from it. Here, mean = 0, variance = finite valueHere, you can see 0 at the center is the Normal Distribution for different mean and variance values. Here is a code example showing the use of Normal Distribution: from scipy.stats import norm  import matplotlib.pyplot as mpl  import numpy as np  def normalDist() -> None:      fig, ax = mpl.subplots(1, 1)      mean, var, skew, kurt = norm.stats(moments = 'mvsk')      x = np.linspace(norm.ppf(0.01),  norm.ppf(0.99), 100)      ax.plot(x, norm.pdf(x),          'r-', lw = 5, alpha = 0.6, label = 'norm pdf')      ax.plot(x, norm.cdf(x),          'b-', lw = 5, alpha = 0.6, label = 'norm cdf')      vals = norm.ppf([0.001, 0.5, 0.999])      np.allclose([0.001, 0.5, 0.999], norm.cdf(vals))      r = norm.rvs(size = 1000)      ax.hist(r, normed = True, histtype = 'stepfilled', alpha = 0.2)      ax.legend(loc = 'best', frameon = False)      mpl.show()  normalDist() Output: Bernoulli Distribution: It is the simplest type of probability distribution. It is a particular case of Binomial distribution, where n=1. It means a binomial distribution takes 'n' number of trials, where n > 1 whereas, the Bernoulli distribution takes only a single trial.   Probability Mass Function of a Bernoulli’s Distribution is:  where p = probability of success and q = probability of failureHere is a code example showing the use of Bernoulli Distribution: from scipy.stats import bernoulli  import seaborn as sb    def bernoulliDist():      data_bern = bernoulli.rvs(size=1200, p = 0.7)      ax = sb.distplot(          data_bern,           kde = True,           color = 'g',           hist_kws = {'alpha' : 1},          kde_kws = {'color': 'y', 'lw': 3, 'label': 'KDE'})      ax.set(xlabel = 'Bernouli Values', ylabel = 'Frequency Distribution')  bernoulliDist() Output:Continuous Uniform Distribution: In this type of continuous distribution, all outcomes are equally possible; each variable gets the same probability of hit as a consequence. This symmetric probabilistic distribution has random variables at an equal interval, with the probability of 1/(b-a). Here is a code example showing the use of Uniform Distribution: from numpy import random  import matplotlib.pyplot as mpl  import seaborn as sb  def uniformDist():      sb.distplot(random.uniform(size = 1200), hist = True)      mpl.show()  uniformDist() Output: Log-Normal Distribution: A Log-Normal distribution is another type of continuous distribution of logarithmic values that form a normal distribution. We can transform a log-normal distribution into a normal distribution. Here is a code example showing the use of Log-Normal Distribution import matplotlib.pyplot as mpl  def lognormalDist():      muu, sig = 3, 1      s = np.random.lognormal(muu, sig, 1000)      cnt, bins, ignored = mpl.hist(s, 80, normed = True, align ='mid', color = 'y')      x = np.linspace(min(bins), max(bins), 10000)      calc = (np.exp( -(np.log(x) - muu) **2 / (2 * sig**2))             / (x * sig * np.sqrt(2 * np.pi)))      mpl.plot(x, calc, linewidth = 2.5, color = 'g')      mpl.axis('tight')      mpl.show()  lognormalDist() Output: Pareto Distribution: It is one of the most critical types of continuous distribution. The Pareto Distribution is a skewed statistical distribution that uses power-law to describe quality control, scientific, social, geophysical, actuarial, and many other types of observable phenomena. The distribution shows slow or heavy-decaying tails in the plot, where much of the data reside at its extreme end. Here is a code example showing the use of Pareto Distribution – import numpy as np  from matplotlib import pyplot as plt  from scipy.stats import pareto  def paretoDist():      xm = 1.5        alp = [2, 4, 6]       x = np.linspace(0, 4, 800)      output = np.array([pareto.pdf(x, scale = xm, b = a) for a in alp])      plt.plot(x, output.T)      plt.show()  paretoDist() Output:Exponential Distribution: It is a type of continuous distribution that determines the time elapsed between events (in a Poisson process). Let’s suppose, that you have the Poisson distribution model that holds the number of events happening in a given period. We can model the time between each birth using an exponential distribution.Here is a code example showing the use of Pareto Distribution – from numpy import random  import matplotlib.pyplot as mpl  import seaborn as sb  def expDist():      sb.distplot(random.exponential(size = 1200), hist = True)      mpl.show()   expDist()Output:Types of the Discrete probability distribution – There are various types of Discrete Probability Distribution a Data science aspirant should know about. Some of them are – Binomial Distribution: It is one of the popular discrete distributions that determine the probability of x success in the 'n' trial. We can use Binomial distribution in situations where we want to extract the probability of SUCCESS or FAILURE from an experiment or survey which went through multiple repetitions. A Binomial distribution holds a fixed number of trials. Also, a binomial event should be independent, and the probability of obtaining failure or success should remain the same. Here is a code example showing the use of Binomial Distribution – from numpy import random  import matplotlib.pyplot as mpl  import seaborn as sb    def binomialDist():      sb.distplot(random.normal(loc = 50, scale = 6, size = 1200), hist = False, label = 'normal')      sb.distplot(random.binomial(n = 100, p = 0.6, size = 1200), hist = False, label = 'binomial')      plt.show()    binomialDist() Output:Geometric Distribution: The geometric probability distribution is one of the crucial types of continuous distributions that determine the probability of any event having likelihood ‘p’ and will happen (occur) after 'n' number of Bernoulli trials. Here 'n' is a discrete random variable. In this distribution, the experiment goes on until we encounter either a success or a failure. The experiment does not depend on the number of trials. Here is a code example showing the use of Geometric Distribution – import matplotlib.pyplot as mpl  def probability_to_occur_at(attempt, probability):      return (1-p)**(attempt - 1) * probability  p = 0.3  attempt = 4  attempts_to_show = range(21)[1:]  print('Possibility that this event will occur on the 7th try: ', probability_to_occur_at(attempt, p))  mpl.xlabel('Number of Trials')  mpl.ylabel('Probability of the Event')  barlist = mpl.bar(attempts_to_show, height=[probability_to_occur_at(x, p) for x in attempts_to_show], tick_label=attempts_to_show)  barlist[attempt].set_color('g')  mpl.show() Output:Poisson Distribution: Poisson distribution is one of the popular types of discrete distribution that shows how many times an event has the possibility of occurrence in a specific set of time. We can obtain this by limiting the Bernoulli distribution from 0 to infinity. Data analysts often use the Poisson distributions to comprehend independent events occurring at a steady rate in a given time interval. Here is a code example showing the use of Poisson Distribution from scipy.stats import poisson  import seaborn as sb  import numpy as np  import matplotlib.pyplot as mpl  def poissonDist():       mpl.figure(figsize = (10, 10))      data_binom = poisson.rvs(mu = 3, size = 5000)      ax = sb.distplot(data_binom, kde=True, color = 'g',                       bins=np.arange(data_binom.min(), data_binom.max() + 1),                       kde_kws={'color': 'y', 'lw': 4, 'label': 'KDE'})      ax.set(xlabel = 'Poisson Distribution', ylabel='Data Frequency')      mpl.show()      poissonDist() Output:Multinomial Distribution: A multinomial distribution is another popular type of discrete probability distribution that calculates the outcome of an event having two or more variables. The term multi means more than one. The Binomial distribution is a particular type of multinomial distribution with two possible outcomes - true/false or heads/tails. Here is a code example showing the use of Multinomial Distribution – import numpy as np  import matplotlib.pyplot as mpl  np.random.seed(99)   n = 12                      pvalue = [0.3, 0.46, 0.22]     s = []  p = []     for size in np.logspace(2, 3):      outcomes = np.random.multinomial(n, pvalue, size=int(size))        prob = sum((outcomes[:,0] == 7) & (outcomes[:,1] == 2) & (outcomes[:,2] == 3))/len(outcomes)      p.append(prob)      s.append(int(size))  fig1 = mpl.figure()  mpl.plot(s, p, 'o-')  mpl.plot(s, [0.0248]*len(s), '--r')  mpl.grid()  mpl.xlim(xmin = 0)  mpl.xlabel('Number of Events')  mpl.ylabel('Function p(X = K)') Output:Negative Binomial Distribution: It is also a type of discrete probability distribution for random variables having negative binomial events. It is also known as the Pascal distribution, where the random variable tells us the number of repeated trials produced during a specific number of experiments.  Here is a code example showing the use of Negative Binomial Distribution – import matplotlib.pyplot as mpl   import numpy as np   from scipy.stats import nbinom    x = np.linspace(0, 6, 70)   gr, kr = 0.3, 0.7        g = nbinom.ppf(x, gr, kr)   s = nbinom.pmf(x, gr, kr)   mpl.plot(x, g, "*", x, s, "r--") Output: Apart from these mentioned distribution types, various other types of probability distributions exist that data science professionals can use to extract reliable datasets. In the next topic, we will understand some interconnections & relationships between various types of probability distributions. Relationship between various Probability distributions – It is surprising to see that different types of probability distributions are interconnected. In the chart shown below, the dashed line is for limited connections between two families of distribution, whereas the solid lines show the exact relationship between them in terms of transformation, variable, type, etc. Conclusion  Probability distributions are prevalent among data analysts and data science professionals because of their wide usage. Today, companies and enterprises hire data science professionals in many sectors, namely, computer science, health, insurance, engineering, and even social science, where probability distributions appear as fundamental tools for application. It is essential for Data analysts and data scientists. to know the core of statistics. Probability Distributions perform a requisite role in analyzing data and cooking a dataset to train the algorithms efficiently. If you want to learn more about data science - particularly probability distributions and their uses, check out KnowledgeHut's comprehensive Data science course. 
9688
Types of Probability Distributions Every Data Scie...

Data Science has become one of the most popular in... Read More