train_neural_bot.py

import numpy as np
from simple_classifier import SimpleClassifier

def vectorize(x):
    # vectorize a string
    if len(x) > 1:
        return np.sum([vectorize(c) for c in x], axis=0)
    if x == '.':
        i = 27
    elif x == ' ':
        i = 26
    else:
        x = x.lower()
        i = ord(x) - 97
    oh = np.zeros(28)
    oh[i] = 1
    return oh

def decode(x):
    # decode a one hot to a character
    i = np.argmax(x) 
    if i == 27:
        return '.'
    if i == 26:
        return ' '
    return chr(i + 97)

# the data
X = ['hi', 'his', 'hise', 'hisen', 'isend', 'send ', 'end n', 'nd nu', 'd nud', ' nude', 'nudes']
Y = ['s', 'e', 'n', 'd', ' ', 'n', 'u', 'd', 'e', 's', '.']

# vectorize data
X = np.array(list(map(vectorize, X)))
Y = np.array(list(map(vectorize, Y)))



clf = SimpleClassifier(28, 28, num_pop=50)
clf.fit(X, Y, epochs=1000, batch_size=11, validation_data=(X, Y))

assert clf.evaluate(X, Y)['accuracy'] == 1

# compress weights
# remove unnecessary weights to obtain sparse matrix
sparse_matrix = []
w = clf.weights[0]
for i in range(28):
    for j in range(28):
        x = w[i, j]
        w[i, j] = 0
        if clf.evaluate(X, Y)['accuracy'] < 1:
            w[i, j] = x
            sparse_matrix.append((i, j, float(str(np.around(x, 1)))))


print(sparse_matrix)

# test
w = np.zeros((28, 28))
for r in sparse_matrix:
    w[r[0], r[1]] = r[2]
clf.weights = [w]
print(clf.evaluate(X, Y))