Hi @Abhinavfreecodecamp In word_index we have values for < START > and < End >.

from grammarcorrection.

can you elaborate more ? with example

from grammarcorrection.

I don't get first you store words in word then you replace it with index of those words, Then you again enumerate over those words ?

from grammarcorrection.

@hadifar please take a look below.
output:
Enter a sentence to correct typo:
hi
your input: hi
hi
746
746
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

code:
#!/usr/bin/env python

-- coding: utf-8 --

import argparse

import numpy as np

import config
import seq2seq_attention
import general_helper

parser = argparse.ArgumentParser()
parser.add_argument('--cache_dir', default="../data/",
help='Path to cache files')

parser.add_argument('--weights_path', default="../models/weights/KerasAttentionNMT.h5",
help='Path to Weights checkpoint')

args = parser.parse_args()

word_index = np.load(open(args.cache_dir + config.CACHE_WORD_INDEX, 'rb'))
embedding = general_helper.load_embedding_matrix(word_index.item())
word_index = word_index.flatten()[0]
index_word = dict([(value, key) for (key, value) in word_index.items()])

model = seq2seq_attention.getModel(embedding, word_index)

model.load_weights(args.weights_path)

def predict(sentence):
words = sentence.split(' ')
#words = word_index[''] + words + word_index['']
words_id = []

for w in words:
    if w in word_index:
        print(w)
        print(word_index[w])
        words_id.append(word_index[w])
    else:
        words_id.append(word_index['<UNK>'])
words = words_id

ret = ""

m_input = [np.zeros((1, config.MAX_SEQ_LEN)), np.zeros((1, config.MAX_SEQ_LEN))]

for i, w in enumerate(words):
    m_input[0][0, i] = w
    print(w)
#m_input[1][0, 0] = word_index['<START>']

for w_i in range(1, config.MAX_SEQ_LEN):
    out = model.predict(m_input)
    out_w_i = out[0][w_i - 1].argmax()
    print(out_w_i)
    if out_w_i == 0:
        continue

    ret += index_word[out_w_i] + " "
    m_input[1][0, w_i] = out_w_i

return ret

while True:
print ("Enter a sentence to correct typo: ")
sent = input()
print ('your input: ' + sent)
print (predict(sent))
print(20 * '-')

from grammarcorrection.

It's teacher forcing mechanism. Take a look at here

from grammarcorrection.

@hadifar it is giving error that ['<'START'>'] in word_index key not found

from grammarcorrection.

@hadifar It would be great if you provide solution as your predict.py code is not working

from grammarcorrection.

Hi @Abhinavfreecodecamp, I really sorry for late response. Unfortunately I am busy these days. < start > keyword must be present in word_index because we added to each sentence during training. I try to deploy the model on Docker to resolve the issue.

from grammarcorrection.

@hadifar Thanks Please Let me know the output

from grammarcorrection.

@hadifar Hi Could you please revert back ASAP

from grammarcorrection.

@hadifar I got the solution, due to version change fit_on_text converts chars to lower case so is actually coded as .

from grammarcorrection.