def test_explain_returns_explanation_with_intercept(self):
"""Tests if the explanation contains an intercept value."""
def _predict_fn(sentences):
return np.random.uniform(0., 1., [len(list(sentences)), 2])
explanation = limsse.explain('Test sentence',
_predict_fn,
1,
num_samples=5)
self.assertNotEqual(explanation.intercept, 0.)
def test_explain_returns_explanation_with_prediction(self):
"""Tests if the explanation contains a prediction."""
def _predict_fn(sentences):
return np.random.uniform(0., 1., [len(list(sentences)), 2])
explanation = limsse.explain('Test sentence',
_predict_fn,
class_to_explain=1,
num_samples=5,
return_prediction=True)
self.assertIsNotNone(explanation.prediction)
def test_explain(self, sentence, num_samples, positive_token,
negative_token, ngram_min_length, ngram_max_length,
num_classes, class_to_explain):
"""Tests explaining a binary classifier with scalar output."""
def _predict_fn(sentences):
"""Mock prediction function."""
predictions = []
for sentence in sentences:
probs = np.random.uniform(0., 1., num_classes)
# To check if LIMSSE finds the right positive/negative correlations.
if negative_token in sentence:
probs[class_to_explain] = probs[class_to_explain] - 1.
if positive_token in sentence:
probs[class_to_explain] = probs[class_to_explain] + 1.
predictions.append(probs)
predictions = np.stack(predictions, axis=0)
if num_classes == 1:
predictions = np.squeeze(special.expit(predictions), -1)
else:
predictions = special.softmax(predictions, axis=-1)
return predictions
explanation = limsse.explain(sentence,
_predict_fn,
class_to_explain,
ngram_min_length=ngram_min_length,
ngram_max_length=ngram_max_length,
num_samples=num_samples,
tokenizer=str.split)
self.assertLen(explanation.feature_importance, len(sentence.split()))
# The positive word should have the highest attribution score.
positive_token_idx = sentence.split().index(positive_token)
self.assertEqual(positive_token_idx,
np.argmax(explanation.feature_importance))
# The negative word should have the lowest attribution score.
negative_token_idx = sentence.split().index(negative_token)
self.assertEqual(negative_token_idx,
np.argmin(explanation.feature_importance))