def do_evaluate(args):
"""
Evaluate an existing model.
"""
logging.info("Evaluating the model.")
model = get_model_factory(args.model).load(args.model_path)
data = list(process_snli_data(args.eval_data))
X1, X2, Y = vectorize_data(data, args.input_length)
emb = WordEmbeddings()
cm = ConfusionMatrix(LABELS)
writer = csv.writer(args.output, delimiter="\t")
writer.writerow(["sentence1", "sentence2", "gold_label", "guess_label", "neutral", "contradiction", "entailment"])
for batch in tqdm(grouper(args.batch_size, zip(data, X1, X2, Y)), total=int(len(data)/args.batch_size)):
objs, X1_batch, X2_batch, y_batch = zip(*batch)
X1_batch = array([emb.weights[x,:] for x in X1_batch])
X2_batch = array([emb.weights[x,:] for x in X2_batch])
y_batch = array(y_batch)
y_batch_ = model.predict_on_batch([X1_batch, X2_batch])
for obj, y, y_ in zip(objs, y_batch, y_batch_):
label = np.argmax(y)
label_ = np.argmax(y_)
writer.writerow([
obj.sentence1,
obj.sentence2,
LABELS[label],
LABELS[label_],
] + list(y_))
cm.update(label, label_)
cm.print_table()
cm.summary()
logging.info("Done.")
def evaluate(args, emb, model, X1X2Y, total=None):
cm = ConfusionMatrix(LABELS)
for batch in tqdm(grouper(args.batch_size, X1X2Y), total=int(total/args.batch_size)):
X1_batch, X2_batch, y_batch = zip(*batch)
X1_batch = array([emb.weights[x,:] for x in X1_batch])
X2_batch = array([emb.weights[x,:] for x in X2_batch])
y_batch = array(y_batch)
y_batch_ = model.predict_on_batch([X1_batch, X2_batch])
for y, y_ in zip(y_batch, y_batch_): cm.update(np.argmax(y), np.argmax(y_))
cm.print_table()
cm.summary()
return cm
def evaluate_prediction(self, session, batch_size, dataset):
print("\nEVALUATING")
cm = ConfusionMatrix(labels=self.LBLS)
total_loss = 0
total_correct = 0
num_batches = 0
for batch in minibatches(dataset, batch_size, bucket=self.bucket):
probs, loss = self.predict(session, batch_size, batch)
_, _, _, _, goldlabels = batch
for i in xrange(len(probs)):
total_correct += 1 if label_to_name(probs[i]) == label_to_name(
goldlabels[i]) else 0
gold_idx = np.argmax(goldlabels[i])
predicted_idx = np.argmax(probs[i])
cm.update(gold_idx, predicted_idx)
total_loss += loss
num_batches += 1
accuracy = total_correct / float(len(dataset[0]))
print("Accuracy: " + str(accuracy))
average_loss = total_loss / float(num_batches)
print("Average Loss: " + str(average_loss))
print("Token-level confusion matrix:\n" + cm.as_table())
print("Token-level scores:\n" + cm.summary())
return (accuracy, average_loss, cm)
def evaluate(model, X, Y):
cm = ConfusionMatrix(labels=LBLS)
Y_ = model.predict(X)
for i in range(Y.shape[0]):
y, y_ = np.argmax(Y[i]), np.argmax(Y_[i])
cm.update(y, y_)
cm.print_table()
return cm.summary()
def evaluate(model, X, Y):
cm = ConfusionMatrix(labels=LBLS)
Y_ = model.predict(X)
for i in range(Y.shape[0]):
y, y_ = np.argmax(Y[i]), np.argmax(Y_[i])
cm.update(y,y_)
cm.print_table()
return cm.summary()
def train(args, emb, model, X1X2Y, total=None):
"""
Train the model using the embeddings @emb and input data batch X1X2Y.
"""
cm = ConfusionMatrix(LABELS)
scorer = Scorer(model.metrics_names)
for batch in tqdm(grouper(args.batch_size, X1X2Y), total=int(total/args.batch_size)):
X1_batch, X2_batch, y_batch = zip(*batch)
X1_batch = array([emb.weights[x,:] for x in X1_batch])
X2_batch = array([emb.weights[x,:] for x in X2_batch])
y_batch = array(y_batch)
score = model.train_on_batch([X1_batch, X2_batch], y_batch)
scorer.update(score, len(y_batch))
y_batch_ = model.predict_on_batch([X1_batch, X2_batch])
for y, y_ in zip(y_batch, y_batch_): cm.update(np.argmax(y), np.argmax(y_))
logging.info("train error: %s", scorer)
cm.print_table()
cm.summary()
return cm
