Introduction to deep learning

Predicting CTR with Machine Learning in Python

Kevin Huo

Instructor

Perceptrons

Example of structure of a perceptron

Input features are standardized
Inputs get summed through weights
Output goes through activation function
Unit step function to convert output into predicted class

Hidden layers and activation functions

Example of hidden layers

Example of activation functions

Implementation

clf = MLPClassifier()
print(clf)

MLPClassifier(activation='relu', 
              alpha=0.0001,

              ... 
              hidden_layer_sizes=(100,),

              learning_rate = 'constant',

              ... 
              max_iter=200, 
              ...)

Other considerations

Standardization is important before usage
- X = StandardScaler().fit_transform(X)
Very large networks with many millions of parameters
- Feature matrices are often "sparse"
Better performance with more data
- However, downside is less transparency and longer compute time

Let's practice!

Predicting CTR with Machine Learning in Python