Classification
Binary Classification results in a decision that is either true or false.
Binary classification Examples:
- Classify whether a medical case is positive or negative, or whether an image contains a hotdog or not a hotdog.
- Classify whether an email is spam or not spam.
Multi-class classification categorizes examples in one of several potential categories (always three or more).
Multi-class classification Examples:
- Classifying whether air quality is poor, moderate, or severe.
- Classifying text into topics such as sports, entertainment, politics, and so on.
Cross-Entropy Loss
Cross-entropy is a score that summarizes the average difference between the actual and predicted probability distributions for all classes. In a classification model, the goal is to minimize the score, with a perfect cross-entropy value is 0.
We can calculate cross-entropy loss by using the log_loss() function in scikit-learn.
# example implementation of cross-entropy loss
true_labels = [1, 0, 0]
predicted_labels = [0.7, 0.2, 0.1]
print(log_loss(true_labels, predicted_labels))Preparing Your Data
To prepare data for cross-entropy loss analysis, you can use the to_categorical() function in TensorFlow’s Keras API to convert labels into one-hot-encodings.
updated_y_train = tensorflow.keras.utils.to_categorical(y_train, dtype = 'int64')
updated_y_test = tensorflow.keras.utils.to_categorical(y_test, dtype = 'int64')F1-Score
In a deep learning classification model, an F1-score can be used to evaluate how our model performs based on how poorly it makes false negative mistakes.
In the code snippet shown, we do the following:
- predict classes for all test cases
my_testusing the scikit-learn.predict()method and assign the result to theyhat_classesvariable. - convert the one-hot-encoded labels
my_test_labelsinto the index of the class the sample belongs to using.argmax()from the NumPy library. The index corresponds to our class encoded as an integer. - use the
.classification_report()method from the scikit-learn library to calculate all the metrics.
import numpy as np
from sklearn.metrics import classification_report
yhat_classes = np.argmax(my_model.predict(my_test), axis = -1)
y_true = np.argmax(my_test_labels, axis=1)
print(classification_report(y_true, yhat_classes))