Principles of Machine Learning

Question 1

Good to remember

Answer 1

In general when doing machine learning, we need to represent each observation in the training and test sets as a vector of numbers, and the label is also represented by a number. ( This is also for classifying images)

Question 2

What is a feature?

Answer 2

Feature. A symbolic or numeric property of a real world object that might be useful to determine its class. The word ‘attribute’ is used for this as well. Different objects however may have different numbers of attributes, while usually for all objects in the same problem the same features can be measured. Thereby objects may be represented by a feature vector, or by a set of attributes.

Question 3

When do you use the Classification algorithm?

Answer 3

Classification. When the data are being used to predict a category, supervised learning is also called classification.

This is the case when assigning an image as a picture of either a ‘cat’ or a ‘dog’. When there are only two choices, this is called two-class or binomial classification. When there are more categories, as when predicting the winner of the NCAA March Madness tournament, this problem is known as multi-class classification.

Question 4

Machine Learning Cheat Sheet : When to use which algorithm?

Answer 4

Question 5

Loss functions for Classification

Question 6

Statistical Learning Theory

Question 7

The key principle in statistical learning theory is the principle of———————————-

Answer 7

Ockham’s razor.

Question 8

What is the curse of dimensionality?

Best Explanations here :

https://www.quora.com/What-is-the-curse-of-dimensionality

Answer 8

The curse of dimensionality refers to how certain learning algorithms may perform poorly in high-dimensional data.

Let’s say you have a straight line 100 yards long and you dropped a penny somewhere on it. It wouldn’t be too hard to find. You walk along the line and it takes two minutes.

Now let’s say you have a square 100 yards on each side and you dropped a penny somewhere on it. It would be pretty hard, like searching across two football fields stuck together. It could take days.

Now a cube 100 yards across. That’s like searching a 30-story building the size of a football stadium. Ugh.

Question 9

What is regularization?

Answer 9

In Machine learning and statistics, a common task is to fit a model to a set of training data. This model can be used later to make predictions or classify new data points.

When the model fits the training data but does not have a good predicting performance and generalization power, we have an overfitting problem.

Regularization is a technique used to avoid this overfitting problem. The idea behind regularization is that models that overfit the data are complex models that have for example too many parameters

Question 10

Logistic Regression

Question 11

How do you use categorical features with scikit-learn?

Answer 11

The categorical features have to be converted to dummy or in python speak in other times called indicator variables.

So there is no difference between a dummy and an indicator variable.

Question 12

How do you calculate Miscalssification Error

Answer 12

(FP + FN ) / N

Question 13

Creating a Classifier in Azure ML

Answer 13

https://courses.edx.org/courses/course-v1:Microsoft+DAT203.2x+6T2016/courseware/786e3fc9eacf4756b3a8091b4a618b4c/c6166f9b4a4641598cc43bab88403a64/

Question 14

What are some of the ways to tune a model?

Answer 14

1. Feature Selection
2. Realization
3. Hyperparameters
4. Cross Validation

Question 15

Come up with stuff for Support VectorMAchines

Question 16

Confusion Matrix : Accuracy,Recall,Precision and F1 Score

Answer 16

Question 17

Which three of the following are reasons for visualizing a data set before attempting to build a supervised machine learning model?

Answer 17

• Develop an understanding of the relationship between the features and the label to determine which features are likely to be predictive of the label and should be used in training the machine learning model.
• Develop an understanding of which features are redundant or collinear with other features and should be eliminated from the dataset before training the machine learning model.
• Find features that are not likely to be predictive of the label and should be removed from the dataset before training the machine learning model.

Question 18

You are training a supervised machine learning model. You want to ensure that when training and testing the dataset you do not introduce any unintentional bias.

What should you do?

Answer 18

Split the dataset into two non-overlapping portions, then train the model using one portion and test it using the other. correct

Question 19

You are training a supervised machine learning model.

Which three kinds of features should be pruned from the dataset?

Answer 19

• Features that are collinear or codependent on other features in the dataset.
• Features that increase model error during training and testing.
• Features that have little impact on model performance during training and testing.

Question 20

While evaluating the performance of a regression model you discover that the residuals are randomly distributed and exhibit no significant structure with respect to the values of the label or the features.

This indicates which two of the following conditions are true?

Answer 20

• The model is likely a good fit to the data.
• The information in the features is being exploited for the most part.

Question 21

When examining the results of cross validation for a machine learning model you notice that the following conditions are true:

1. The values of the metrics are similar across the folds.
2. The standard deviation of the metrics is small compared to the mean values.
3. The mean values of the metrics are in an acceptable range.

Given these conditions you can conclude that which of the following statements is true?

Answer 21

The performance of the model should generalize well

Question 22

While exploring a dataset for training a two-class (binary) classification problem, you notice the following properties of the data set:

• Certain features exhibit noticeable separation in the values or categories based on the categories of the label.
• Certain features exhibit little separation in the values or categories based on the categories of the label.

Which two of the following feature selection actions should you take before training the model?

Answer 22

• The features exhibiting separation should be retained in the dataset.
• The features with poor separation should be pruned from the dataset.

Question 23

When exploring the k-means clustering of a data set you examine the projections of the first two principle component of the cluster ellipses, for a certain number of clusters, and you observe:

• The major and minor axes of each of the ellipses are of distinctly different length.
• The directions of major axes of the ellipses are distinctly different.

Which of the following statements is true?

Answer 23

The number of clusters chosen fits this dataset well.

Question 24

You create an experiment that uses a Train Matchbox Recommender module to train a recommendation model, and add a Score Matchbox Recommender module to generate a prediction. You want to use the model in an online retail site to predict additional items a user might want to purchase based on the item they are currently viewing.

Which recommender prediction kind should you configure the Score Matchbox Recommender module to use?

Answer 24

Related Items .

Question 25

You create an experiment that uses a Train Matchbox Recommender module and a Score Matchbox Recommender module with the recommender prediction kind set to Item Recommendation.

How should you evaluate the performance of your model?

Answer 25

Add an Evaluate Recommender module and review the Normalized Discounted Cumulative Gain (NDCG) value. The closer this value is to 1, the better the model is performing.

Question 26

You have trained and tested a regression model, and plotted the predicted labels against the actual label values in the training and test data. The resulting chart shows strong correlation between the predicted labels and the training data, but the actual label values from the test data vary significantly from the predicted values.

What can you conclude about your model?

Answer 26

The model is over-fitted.

Question 27

You create an Azure ML experiment for a dataset of credit card transactions.

The experiment contains two classifications models: One using a Two-Class Decision Forest algorithm and one using a Two-Class Support Vector Machine algorithm.

When you evaluate the models, they exhibit the following performance metrics:

• Two-Class Decision Forest: Accuracy = 0.964, Recall = 0.945, AUC = 0.987
• Two-Class Support Vector Machine: Accuracy = 0.765, Recall = 0.768, AUC=0.799

What should you conclude from these metrics?

Answer 27

For this dataset, the Two-Class Decision Forest model appears to be the best model.

Question 28

You are preparing data for a classification model. The dataset you are using includes a large number of columns, not all of which are useful in predicting the label. You have used the Permutation Feature Importance module in Azure ML to determine the statistical importance of each feature in the dataset.

What should you do next?

Answer 28

Iteratively remove the least important features until the model accuracy starts to degrade.

Question 29

You are creating a classification model using the Two-Class Logistical Regression algorithm. You want to identify the optimal values for the regularization weight parameters for this algorithm.

What should you do?

Answer 29

Split the test dataset and use one portion to try a range of parameters with the Tune Model Hyperparameters module, and another to score the trained model it produces.

Question 30

When exploring the k-means clustering of a dataset you continue to increase the number of clusters one-by-one until you observe that the projection of the first two principal components show:

• The major and minor axes of each of the ellipses are of similar lengths.
• The directions of major axes of the ellipses are relatively aligned.

Based on these observations, what should you do?

Answer 30

Reduce the number of clusters chosen.

Question 31

You create a binary (two-class) classification machine learning model. When evaluating the model, you observe the following metrics:

• Accuracy: 0.9
• Area under the curve (AUC): 0.8
• Recall: 0.3

What can you conclude about the performance of your model?

Answer 31

Proportionately, false negatives predominate the errors.

Question 32

You create a regression model, and then cross-validate it.

You observe the following results:

• The values of the performance metrics change significantly from fold to fold.
• The standard deviation of the performance metrics is close to the value of the metrics.

What can you conclude about your mode?

Answer 32

The model will likely not generalize well.

Question 33

You create an experiment that uses a Train Matchbox Recommender module to train a recommendation model, and add a Score Matchbox Recommender module to generate a prediction. You want to use the model in a music streaming service to recommend songs for the currently logged in user.

Which recommender prediction kind should you configure the Score Matchbox Recommender module to use?

Answer 33

Item Recommendation

Question 34

You have created two models in Azure ML, using two different classification algorithms.

The following ROC chart shows the relative performance of these models, with Model A represented as theScored dataset and Model B represented as the Scored dataset to compare.

What can you determine from the ROC chart?

Answer 34

Model A performs better than Model B

Question 35

Feature Selection : Good to Remember

Answer 35

Selecting the best features is essential to the optimal performance of machine learning models.

Only features that contribute to measurably improving model performance should be used.
Using extraneous features can contribute noise to training and predictions from machine learning models. This behavior can prevent machine learning models from generalizing from training data to data received in production.

Question 36

GEnerally how is feature selection done?

Answer 36

Feature selection is generally performed in one of two ways.

• In forward stepwise feature selection, the most promising features are added one at a time.
• In reverse stepwise feature selection, the least

important features are pruned one, or a few, at a time.

Question 37

Which module in Azure ML measures the relationship between the features and the label and ranks their importance ( co-efficients)?

Answer 37

Permutation Feature Importance Module.

Note: When examining the results produced by the Permutation Feature Importance module, the
absolute values of the feature importance measure should be compared.

Question 38

What is Regularization?

Answer 38

Regularization is an alternative to feature selection. In summary, regularization forces the weights of less important features toward zero. This process helps to ensure that models will generalize in
production. Regularization can be performed on models with very large numbers of features, for which manual feature selection is infeasible.

Question 39

What is Cross Validation?

Answer 39

Cross validation uses a resampling scheme to train and evaluate machine learning model. The model is trained and evaluated with a series of folds, created by resampling with replacement. The model performance metrics are averaged over the results of the folds. This aggregated performance measure is more likely to represent the actual performance a model will achieve in production.

Question 40

What can Ada-boosted models be used for?

Answer 40

Ada-boosted tree models can be used for classification or regression.

Question 41

How to Determine the Optimal Number of Clusters in Azure ML?

Answer 41

K-Means Clustering module

Question 42

Good to Remember : Recommenders

Answer 42

Recommenders are an interesting and useful class of machine learning models. Creating good recommenders is challenging since there is no objective way to measure how good a recommendation is for a given individual. There is no way to know if the recommendation is the best possible for an individual. Further, the ratings provided by the users, is based on their personal subjective judgement.

Question 43

What does AUC stand for?

Answer 43

Area Under Curve