Path9.Mod1.b - Selecting Multi-Classification Algorithms for Azure ML

Question 1

When you want to Predict between several categories with fast training times resulting in a linear model

Answer 1

Multiclass Logistic Regression … extends LR to determine probabilities of each class using Softmax, then selects the best one

Question 2

When you want to Predict between several categories with fast training times, using a non-parametric algorithm that scales well and memory consumption isn’t a concern

Answer 2

Multiclass Boosted Decision Tree

Question 3

When you want to Predict between several categories where Accuracy is the priority and training time is not a concern

Answer 3

Multiclass Neural Network

Question 4

When you want to Predict between several categories where accuracy, training time and the ability to control memory consumption are the priorities

Answer 4

Multiclass Decision Forest

Question 5

When you want to Predict between several categories where your data has a large number of classes but not a lot of data and you need a simple approach to classifying that lends well to parallelism when training

Explain how the algorithm works

Answer 5

One-vs-All Multiclass. Each class impliments a Binary Classifier to distiguish Class_A (positive) from the remaining Classes_B-n (negative),

Question 6

When you want to Predict between several categories where the data may have several class imbalances but still gets reliably high accuracy

Explain how the algorithm works

Answer 6

One-vs-One Multiclass. All possible pairs of classes impliment a Binary Classifier, and final prediction is made between (N(N-1))/2 total Classifiers where the class picked the most wins