CODEWITHSUNDEEP
pythonscikit-learnperceptron
andreSmol
andreSmol

Reputation: 1038

scikit Perceptron bias

I am using the very basic linear classifier provided by scikit class Perceptron:

  clf = linear_model.Perceptron(n_iter=12)
  clf.fit(X,Y)

I have a X array where the rows are instances and the columns are binary features. I have an Y array with my classes. My data has three classes. I have two questions: 1)The perceptron algorithm requires a bias term. How is the scikit perceptron handling the bias? Should I add an "bias column" (all ones) to my input X data? Or does the scikit perceptron function adds automatically an bias to the X array (input) with features? Or is it handling the bias separatly? 2)How can find the training error for my perceptron?

Upvotes: 2

Views: 2369

Answers (2)

Fred Foo
Fred Foo

Reputation: 363547

  1. The bias terms is learned automatically, and can be found in clf.intercept_ after you've called fit (it's an array with one term per class).

  2. clf.score(X) gives the accuracy on X. 1 - clf.score(X) is the zero-one loss (error).

Upvotes: 2

user1269942
user1269942

Reputation: 3852

1) the bias will be dealt with automatically. If you're unsure, try training with 2 versions of your data...your original data and a scaled version (standard scaler in sklearn).

2)

clf = linear_model.Perceptron(n_iter=12)
clf.fit(X, Y)

training_results = clf.predict(X)
training_error = 1 - metrics.accuracy_score(training_results, Y)  #or pick your metric from metrics module.

As you can see, if you calculate the error when predicting with the data you trained with, this gives the training error. "Test error" is when you predict on data that your model has not "seen" yet. I subtracted from 1 because accuracy gives the percent of successful matches (a measure of success) whereas training error is a measure of error. There are many types of error...accuracy is just one.

Upvotes: 2

Related Questions