tester.py

import tensorflow as tf
from nwp import NWP, NWPV2
import pickle
import numpy as np

predictor = NWP(int(open("num_vocabulary.txt","r").read()))

predictor.load_weights("checkpoints/February-04-2021_03+21PM/best_model")

vocabulary = pickle.load(open("vocabulary.pckl", "rb"))
vocabulary = dict(zip(vocabulary.values(), vocabulary.keys()))


def test_cases(model, test_phrases, vocabulary):
    test_classes = model.inference(test_phrases)
    top_k = tf.math.top_k(test_classes, k=10)
    top_k_probs = top_k[0].numpy()[0]
    top_k_probs = top_k_probs*1.0/np.sum(top_k_probs)
    top_k = top_k[1].numpy()[0]

    idx = np.random.choice(top_k, p=top_k_probs)

    splitted_phrase = test_phrases[0].split(' ')
    print(vocabulary[idx])

    # print([vocabulary[top_k[i]] for i in range(10) if vocabulary[top_k[i]] != splitted_phrase[-1]])

while True:
    phrase = input("Input: ")
    print(test_cases(predictor, [phrase], vocabulary))