from spacy.lang.en import English
from spacy.pipeline import Sentencizer
from keras.preprocessing.text import Tokenizer
from keras.layers import Input, Dense, Embedding, Flatten
from keras.models import Model
from tensorflow.keras.preprocessing.sequence import pad_sequences
import numpy as np
import re

nlp = English()
sentencizer = Sentencizer()
np.set_printoptions(precision=2)

with open("anime.txt", encoding="utf-8") as f:
    anime_text = re.sub(r'\[\d\]','',f.read())
    an = sentencizer(nlp(anime_text))
    anime_sents = [ span.text.strip() for span in an.sents ]

with open("basketball.txt", encoding="utf-8") as f:
    basketball_text = re.sub(r'\[\d\]','',f.read())
    ba = sentencizer(nlp(basketball_text))
    basketball_sents = [ span.text.strip() for span in ba.sents ]

labels = np.concatenate([
    np.repeat(0, len(anime_sents)),
    np.repeat(1, len(basketball_sents)),
    ])

t = Tokenizer()
t.fit_on_texts(anime_sents + basketball_sents)

vocab_size = len(t.word_index) + 1

E = np.zeros(shape=(vocab_size, 50))
with open("glove.6B.50d.txt", encoding='utf-8') as f:
    for line in f:
        stuff = line.split()
        word = stuff[0]
        numbers = np.array([ float(x) for x in stuff[1:] ])
        if word in t.word_index:
            print(f"Yes! adding embedding for word {word} (at row {t.word_index[word]})")
            E[t.word_index[word],:] = numbers


encoded = t.texts_to_sequences(anime_sents + basketball_sents)
padded = pad_sequences(encoded, maxlen=10, padding="post")
input_layer = Input(shape=(padded.shape[1],))
e_layer = Embedding(vocab_size, 50, input_length=10, weights=[E],
    trainable=False)(input_layer)
f_layer = Flatten()(e_layer)
hidden_layer = Dense(20, activation='relu')(f_layer)
output_layer = Dense(1, activation='sigmoid')(hidden_layer)
model = Model(inputs=input_layer, outputs=output_layer)
model.compile(loss='binary_crossentropy', metrics=['accuracy'])
model.fit(padded, labels, epochs=20, verbose=0)


sent = input("Enter a sentence: ")
while (sent != 'done'):
    encoded_sent = t.texts_to_sequences([sent])
    padded_sent = pad_sequences(encoded_sent, maxlen=10, padding="post")
    print(model(padded_sent))
    sent = input("Enter a sentence: ")