-
Notifications
You must be signed in to change notification settings - Fork 0
/
callback.py
42 lines (33 loc) · 1.65 KB
/
callback.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
# Importing dependencies
import keras
import numpy as np
# Importing some constants
from constants import DEMO_REVIEW, PREDICT_CHARS, MAX_LENGTH, TEMPERATURE
# Class CustomCallback
class CustomCallback(keras.callbacks.Callback):
# This method intriduces some randomness in the prediction
def sample(self, preds, temperature=1.0):
preds = np.asarray(preds).astype('float64')
preds = np.log(preds) / temperature
exp_preds = np.exp(preds)
preds = exp_preds / np.sum(exp_preds)
probas = np.random.multinomial(1, preds, 1)
return np.argmax(probas)
# This method runs after each epoch
def on_epoch_end(self, epoch, logs={}):
# Printing some info and predicted text
print('Currently at epoch {}'.format(epoch + 1))
print('Starter text : {}'.format(DEMO_REVIEW))
for temperature in TEMPERATURE:
# Selecting first 40 characters and also converting the text into lowercase as the model supports it only
review = DEMO_REVIEW.lower()[0:MAX_LENGTH]
review = [ord(i) for i in review]
# Predicting the next characters for PREDICT_CHARS times
for k in range(PREDICT_CHARS):
# Predicting using the model
temp = self.model.predict(np.array([review[k: k + MAX_LENGTH]]))
# Calling the sample method
temp = self.sample(temp[0], temperature)
# Appending the predicted charcter
review.append(temp)
print('Generated text with temperature {}: {}'.format(temperature, ''.join([chr(i) for i in review])))