def evaluation(x, y):
y = tf.cast(y, dtype=tf.float32)
y_pred = predict(x)
predict_accuracy = tf.keras.metrics.BinaryAccuracy(name='predict_accuracy')
acc = predict_accuracy(y, y_pred)
mi_f1 = micro_f1(y, y_pred)
ma_f1 = macro_f1(y, y_pred)
print("val accuracy {:.4f}, micro f1 {:.4f} macro f1 {:.4f}".format(
acc.numpy(), mi_f1.numpy(), ma_f1.numpy()))
return acc, mi_f1, ma_f1
def test(model,x_test,y_test):
print('Start Testing......')
y_pred = model.predict(x_test)
y_pred = tf.constant(y_pred,tf.float32)
y_test = tf.constant(y_test,tf.float32)
print(micro_f1(y_test,y_pred))
print(macro_f1(y_test,y_pred))
print(classification_report(y_test,y_pred))
def train_step(x, y):
enc_padding_mask = create_padding_mask(x)
with tf.GradientTape() as tape:
y_pred = model(x, training=True, enc_padding_mask=enc_padding_mask)
loss = loss_object(y, y_pred)
grads = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
train_loss(loss)
train_accuracy(y, y_pred)
mi_f1 = micro_f1(y, y_pred)
ma_f1 = macro_f1(y, y_pred)
return mi_f1, ma_f1, y_pred
def evaluate(test_dataset):
predictions = []
tars = []
for (batch, (inp, tar)) in tqdm(enumerate(test_dataset)):
enc_padding_mask = create_padding_mask(inp)
predict = transformer(inp, False, enc_padding_mask=enc_padding_mask)
predictions.append(predict)
tars.append(tar)
predictions = tf.concat(predictions, axis=0)
tars = tf.concat(tars, axis=0)
mi_f1 = micro_f1(tars, predictions)
ma_f1 = macro_f1(tars, predictions)
predictions = np.where(predictions > 0.5, 1, 0)
tars = np.where(tars > 0.5, 1, 0)
smaple_f1 = f1_score(tars, predictions, average='samples')
return mi_f1, ma_f1, smaple_f1, tars, predictions
def train_step(inp, tar):
enc_padding_mask = create_padding_mask(inp)
with tf.GradientTape() as tape:
predictions = transformer(inp,
training=True,
enc_padding_mask=enc_padding_mask)
loss = loss_function(tar, predictions)
gradients = tape.gradient(loss, transformer.trainable_variables)
optimizer.apply_gradients(zip(gradients, transformer.trainable_variables))
train_loss(loss)
train_accuracy(tar, predictions)
mi_f1 = micro_f1(tar, predictions)
ma_f1 = macro_f1(tar, predictions)
return mi_f1, ma_f1
def predict(inp, tar, enc_padding_mask):
predictions = transformer(inp, False, enc_padding_mask=enc_padding_mask)
mi_f1 = micro_f1(tar, predictions)
ma_f1 = macro_f1(tar, predictions)
return mi_f1, ma_f1