from __future__ import absolute_import, division, print_function, unicode_literals
import tensorflow as tf
import matplotlib.pyplot as pltimport matplotlib.ticker as tickerfrom sklearn.model_selection import train_test_split
import unicodedataimport reimport numpy as npimport osimport ioimport time

Download and prepare the dataset

May I borrow this book? ¿Puedo tomar prestado este libro?

1. Add a start and end token to each sentence.
2. Clean the sentences by removing special characters.
3. Create a word index and reverse word index (dictionaries mapping from word → id and id → word).
4. Pad each sentence to a maximum length.

# Download the file
path_to_zip = tf.keras.utils.get_file(
    'spa-eng.zip', origin='http://storage.googleapis.com/download.tensorflow.org/data/spa-eng.zip',
    extract=True)

path_to_file = os.path.dirname(path_to_zip)+"/spa-eng/spa.txt"

Downloading data from http://storage.googleapis.com/download.tensorflow.org/data/spa-eng.zip
2646016/2638744 [==============================] - 0s 0us/step

# Converts the unicode file to ascii
def unicode_to_ascii(s):
    return ''.join(c for c in unicodedata.normalize('NFD', s)
        if unicodedata.category(c) != 'Mn')

def preprocess_sentence(w):
    w = unicode_to_ascii(w.lower().strip())

    # creating a space between a word and the punctuation following it
    # eg: "he is a boy." => "he is a boy ."
    # Reference:- https://stackoverflow.com/questions/3645931/python-padding-punctuation-with-white-spaces-keeping-punctuation
    w = re.sub(r"([?.!,¿])", r" \1 ", w)
    w = re.sub(r'[" "]+', " ", w)

    # replacing everything with space except (a-z, A-Z, ".", "?", "!", ",")
    w = re.sub(r"[^a-zA-Z?.!,¿]+", " ", w)

    w = w.rstrip().strip()

    # adding a start and an end token to the sentence
    # so that the model know when to start and stop predicting.
    w = '<start> ' + w + ' <end>'
    return w

en_sentence = u"May I borrow this book?"
sp_sentence = u"¿Puedo tomar prestado este libro?"
print(preprocess_sentence(en_sentence))
print(preprocess_sentence(sp_sentence).encode('utf-8'))

<start> may i borrow this book ? <end>
b'<start> \xc2\xbf puedo tomar prestado este libro ? <end>'

# 1. Remove the accents
# 2. Clean the sentences
# 3. Return word pairs in the format: [ENGLISH, SPANISH]
def create_dataset(path, num_examples):
    lines = io.open(path, encoding='UTF-8').read().strip().split('\n')

    word_pairs = [[preprocess_sentence(w) for w in l.split('\t')]  for l in lines[:num_examples]]

    return zip(*word_pairs)

en, sp = create_dataset(path_to_file, None)
print(en[-1])
print(sp[-1])

<start> if you want to sound like a native speaker , you must be willing to practice saying the same sentence over and over in the same way that banjo players practice the same phrase over and over until they can play it correctly and at the desired tempo . <end>
<start> si quieres sonar como un hablante nativo , debes estar dispuesto a practicar diciendo la misma frase una y otra vez de la misma manera en que un musico de banjo practica el mismo fraseo una y otra vez hasta que lo puedan tocar correctamente y en el tiempo esperado . <end>

def max_length(tensor):
    return max(len(t) for t in tensor)

def tokenize(lang):
  lang_tokenizer = tf.keras.preprocessing.text.Tokenizer(
      filters='')
  lang_tokenizer.fit_on_texts(lang)

  tensor = lang_tokenizer.texts_to_sequences(lang)

  tensor = tf.keras.preprocessing.sequence.pad_sequences(tensor,
                                                         padding='post')

  return tensor, lang_tokenizer

def load_dataset(path, num_examples=None):
    # creating cleaned input, output pairs
    targ_lang, inp_lang = create_dataset(path, num_examples)

    input_tensor, inp_lang_tokenizer = tokenize(inp_lang)
    target_tensor, targ_lang_tokenizer = tokenize(targ_lang)

    return input_tensor, target_tensor, inp_lang_tokenizer, targ_lang_tokenizer

Limit the size of the dataset to experiment faster (optional)

# Try experimenting with the size of that dataset
num_examples = 30000
input_tensor, target_tensor, inp_lang, targ_lang = load_dataset(path_to_file, num_examples)
# Calculate max_length of the target tensors
max_length_targ, max_length_inp = max_length(target_tensor), max_length(input_tensor)

# Creating training and validation sets using an 80-20 split
input_tensor_train, input_tensor_val, target_tensor_train, target_tensor_val = train_test_split(input_tensor, target_tensor, test_size=0.2)
# Show length
print(len(input_tensor_train), len(target_tensor_train), len(input_tensor_val), len(target_tensor_val))

24000 24000 6000 6000

def convert(lang, tensor):
  for t in tensor:
    if t!=0:
      print ("%d ----> %s" % (t, lang.index_word[t]))

print ("Input Language; index to word mapping")
convert(inp_lang, input_tensor_train[0])
print ()
print ("Target Language; index to word mapping")
convert(targ_lang, target_tensor_train[0])

Input Language; index to word mapping
1 ----> <start>
22 ----> por
50 ----> favor
456 ----> escucha
3 ----> .
2 ----> <end>

Target Language; index to word mapping
1 ----> <start>
56 ----> please
279 ----> listen
3 ----> .
2 ----> <end>

Create a tf.data dataset

BUFFER_SIZE = len(input_tensor_train)
BATCH_SIZE = 64
steps_per_epoch = len(input_tensor_train)//BATCH_SIZE
embedding_dim = 256
units = 1024
vocab_inp_size = len(inp_lang.word_index)+1
vocab_tar_size = len(targ_lang.word_index)+1

dataset = tf.data.Dataset.from_tensor_slices((input_tensor_train, target_tensor_train)).shuffle(BUFFER_SIZE)
dataset = dataset.batch(BATCH_SIZE, drop_remainder=True)

example_input_batch, example_target_batch = next(iter(dataset))
example_input_batch.shape, example_target_batch.shape

(TensorShape([64, 16]), TensorShape([64, 11]))

 

• FC = Fully connected (dense) layer
• EO = Encoder output
• H = hidden state
• X = input to the decoder
And the pseudo-code:
• score = FC(tanh(FC(EO) + FC(H)))

• attention weights = softmax(score, axis = 1)। ডিফল্টরূপে সফটম্যাক্স শেষ অক্ষটিতে প্রয়োগ করা হয় তবে এখানে আমরা এটি 1 ম অক্ষের উপর প্রয়োগ করতে চাই , যেহেতু স্কোরের আকার (ব্যাচ_সাইজ, সর্বোচ্চ_ দৈর্ঘ্য, গোপন_সাইজ) । Max_lengthআমাদের ইনপুট দৈর্ঘ্য হয়। যেহেতু আমরা প্রতিটি ইনপুটে একটি ওজন নির্ধারণের চেষ্টা করছি, তাই সেই অক্ষের উপর সফটম্যাক্স প্রয়োগ করা উচিত।
• context vector = sum(attention weights * EO, axis = 1)। অক্ষ হিসাবে 1 হিসাবে বেছে নেওয়ার জন্য উপরের একই কারণ।
• embedding output = ডিকোডার এক্স এর ইনপুটটি একটি এম্বেডিং স্তরের মধ্য দিয়ে যায়।
• merged vector = concat(embedding output, context vector)
• এই মার্জ করা ভেক্টরটি তখন জিআরইউতে দেওয়া হয়

class Encoder(tf.keras.Model):
  def __init__(self, vocab_size, embedding_dim, enc_units, batch_sz):
    super(Encoder, self).__init__()
    self.batch_sz = batch_sz
    self.enc_units = enc_units
    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)
    self.gru = tf.keras.layers.GRU(self.enc_units,
                                   return_sequences=True,
                                   return_state=True,
                                   recurrent_initializer='glorot_uniform')

  def call(self, x, hidden):
    x = self.embedding(x)
    output, state = self.gru(x, initial_state = hidden)
    return output, state

  def initialize_hidden_state(self):
    return tf.zeros((self.batch_sz, self.enc_units))

encoder = Encoder(vocab_inp_size, embedding_dim, units, BATCH_SIZE)
# sample input
sample_hidden = encoder.initialize_hidden_state()
sample_output, sample_hidden = encoder(example_input_batch, sample_hidden)
print ('Encoder output shape: (batch size, sequence length, units) {}'.format(sample_output.shape))
print ('Encoder Hidden state shape: (batch size, units) {}'.format(sample_hidden.shape))

class BahdanauAttention(tf.keras.layers.Layer):
  def __init__(self, units):
    super(BahdanauAttention, self).__init__()
    self.W1 = tf.keras.layers.Dense(units)
    self.W2 = tf.keras.layers.Dense(units)
    self.V = tf.keras.layers.Dense(1)

  def call(self, query, values):
    # hidden shape == (batch_size, hidden size)
    # hidden_with_time_axis shape == (batch_size, 1, hidden size)
    # we are doing this to perform addition to calculate the score
    hidden_with_time_axis = tf.expand_dims(query, 1)

    # score shape == (batch_size, max_length, 1)
    # we get 1 at the last axis because we are applying score to self.V
    # the shape of the tensor before applying self.V is (batch_size, max_length, units)
    score = self.V(tf.nn.tanh(
        self.W1(values) + self.W2(hidden_with_time_axis)))

    # attention_weights shape == (batch_size, max_length, 1)
    attention_weights = tf.nn.softmax(score, axis=1)

    # context_vector shape after sum == (batch_size, hidden_size)
    context_vector = attention_weights * values
    context_vector = tf.reduce_sum(context_vector, axis=1)

    return context_vector, attention_weights

attention_layer = BahdanauAttention(10)
attention_result, attention_weights = attention_layer(sample_hidden, sample_output)
print("Attention result shape: (batch size, units) {}".format(attention_result.shape))
print("Attention weights shape: (batch_size, sequence_length, 1) {}".format(attention_weights.shape))

class Decoder(tf.keras.Model):
  def __init__(self, vocab_size, embedding_dim, dec_units, batch_sz):
    super(Decoder, self).__init__()
    self.batch_sz = batch_sz
    self.dec_units = dec_units
    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)
    self.gru = tf.keras.layers.GRU(self.dec_units,
                                   return_sequences=True,
                                   return_state=True,
                                   recurrent_initializer='glorot_uniform')
    self.fc = tf.keras.layers.Dense(vocab_size)

    # used for attention
    self.attention = BahdanauAttention(self.dec_units)

  def call(self, x, hidden, enc_output):
    # enc_output shape == (batch_size, max_length, hidden_size)
    context_vector, attention_weights = self.attention(hidden, enc_output)

    # x shape after passing through embedding == (batch_size, 1, embedding_dim)
    x = self.embedding(x)

    # x shape after concatenation == (batch_size, 1, embedding_dim + hidden_size)
    x = tf.concat([tf.expand_dims(context_vector, 1), x], axis=-1)

    # passing the concatenated vector to the GRU
    output, state = self.gru(x)

    # output shape == (batch_size * 1, hidden_size)
    output = tf.reshape(output, (-1, output.shape[2]))

    # output shape == (batch_size, vocab)
    x = self.fc(output)

    return x, state, attention_weights

decoder = Decoder(vocab_tar_size, embedding_dim, units, BATCH_SIZE)

sample_decoder_output, _, _ = decoder(tf.random.uniform((64, 1)),
                                      sample_hidden, sample_output)
print ('Decoder output shape: (batch_size, vocab size) {}'.format(sample_decoder_output.shape))

অপ্টিমাইজার এবং ক্ষতি ফাংশন সংজ্ঞায়িত করুন

optimizer = tf.keras.optimizers.Adam()
loss_object = tf.keras.losses.SparseCategoricalCrossentropy(
    from_logits=True, reduction='none')
def loss_function(real, pred):
  mask = tf.math.logical_not(tf.math.equal(real, 0))
  loss_ = loss_object(real, pred)

  mask = tf.cast(mask, dtype=loss_.dtype)
  loss_ *= mask

  return tf.reduce_mean(loss_)

চেকপয়েন্টগুলি (অবজেক্ট-ভিত্তিক সঞ্চয়)

checkpoint_dir = './training_checkpoints'
checkpoint_prefix = os.path.join(checkpoint_dir, "ckpt")
checkpoint = tf.train.Checkpoint(optimizer=optimizer,
                                 encoder=encoder,
                                 decoder=decoder)

প্রশিক্ষণ

1. এনকোডার দিয়ে ইনপুটটি পাস করুন যা এনকোডার আউটপুট এবং এনকোডার লুকানো অবস্থায় ফিরে আসে ।
2. এনকোডার আউটপুট, এনকোডার লুকানো অবস্থা এবং ডিকোডার ইনপুট (যা শুরু টোকেন ) ডিকোডারে স্থানান্তরিত হয় to
3. ডিকোডারটি পূর্বাভাসগুলি এবং ডিকোডার লুকানো অবস্থায় ফিরে আসে ।
4. এর পরে ডিকোডার লুকানো অবস্থা আবার মডেলটিতে ফিরে যায় এবং ভবিষ্যদ্বাণীগুলি ক্ষতির গণনা করতে ব্যবহৃত হয়।
5. ডিকোডারটিতে পরবর্তী ইনপুট সিদ্ধান্ত নিতে শিক্ষককে বাধ্য করুন Use
6. শিক্ষক জোরপূর্বক হ'ল কৌশলটি যেখানে লক্ষ্য শব্দটি ডিকোডারের পরবর্তী ইনপুট হিসাবে পাস করা হয় ।
7. চূড়ান্ত পদক্ষেপটি গ্রেডিয়েন্টগুলি গণনা করা এবং এটি অপটিমাইজার এবং ব্যাকপ্রোপগেটে প্রয়োগ করা।

@tf.function
def train_step(inp, targ, enc_hidden):
  loss = 0

  with tf.GradientTape() as tape:
    enc_output, enc_hidden = encoder(inp, enc_hidden)

    dec_hidden = enc_hidden

    dec_input = tf.expand_dims([targ_lang.word_index['<start>']] * BATCH_SIZE, 1)

    # Teacher forcing - feeding the target as the next input
    for t in range(1, targ.shape[1]):
      # passing enc_output to the decoder
      predictions, dec_hidden, _ = decoder(dec_input, dec_hidden, enc_output)

      loss += loss_function(targ[:, t], predictions)

      # using teacher forcing
      dec_input = tf.expand_dims(targ[:, t], 1)

  batch_loss = (loss / int(targ.shape[1]))

  variables = encoder.trainable_variables + decoder.trainable_variables

  gradients = tape.gradient(loss, variables)

  optimizer.apply_gradients(zip(gradients, variables))

  return batch_loss

EPOCHS = 10
for epoch in range(EPOCHS):
  start = time.time()

  enc_hidden = encoder.initialize_hidden_state()
  total_loss = 0

  for (batch, (inp, targ)) in enumerate(dataset.take(steps_per_epoch)):
    batch_loss = train_step(inp, targ, enc_hidden)
    total_loss += batch_loss

    if batch % 100 == 0:
        print('Epoch {} Batch {} Loss {:.4f}'.format(epoch + 1,
                                                     batch,
                                                     batch_loss.numpy()))
  # saving (checkpoint) the model every 2 epochs
  if (epoch + 1) % 2 == 0:
    checkpoint.save(file_prefix = checkpoint_prefix)

  print('Epoch {} Loss {:.4f}'.format(epoch + 1,
                                      total_loss / steps_per_epoch))
  print('Time taken for 1 epoch {} sec\n'.format(time.time() - start))

Epoch 1 Batch 0 Loss 4.5782
Epoch 1 Batch 100 Loss 2.3247
Epoch 1 Batch 200 Loss 1.8299
Epoch 1 Batch 300 Loss 1.7017
Epoch 1 Loss 2.0273
Time taken for 1 epoch 32.79620003700256 sec

Epoch 2 Batch 0 Loss 1.5939
Epoch 2 Batch 100 Loss 1.4579
Epoch 2 Batch 200 Loss 1.4302
Epoch 2 Batch 300 Loss 1.3060
Epoch 2 Loss 1.3856
Time taken for 1 epoch 17.484558582305908 sec

Epoch 3 Batch 0 Loss 0.9645
Epoch 3 Batch 100 Loss 0.9450
Epoch 3 Batch 200 Loss 0.9626
Epoch 3 Batch 300 Loss 1.0414
Epoch 3 Loss 0.9681
Time taken for 1 epoch 17.044427633285522 sec

Epoch 4 Batch 0 Loss 0.6285
Epoch 4 Batch 100 Loss 0.7940
Epoch 4 Batch 200 Loss 0.5498
Epoch 4 Batch 300 Loss 0.6397
Epoch 4 Loss 0.6515
Time taken for 1 epoch 17.429885387420654 sec

Epoch 5 Batch 0 Loss 0.4643
Epoch 5 Batch 100 Loss 0.4660
Epoch 5 Batch 200 Loss 0.4049
Epoch 5 Batch 300 Loss 0.4017
Epoch 5 Loss 0.4392
Time taken for 1 epoch 17.022470474243164 sec

Epoch 6 Batch 0 Loss 0.2925
Epoch 6 Batch 100 Loss 0.2970
Epoch 6 Batch 200 Loss 0.2859
Epoch 6 Batch 300 Loss 0.2650
Epoch 6 Loss 0.3011
Time taken for 1 epoch 17.285199642181396 sec

Epoch 7 Batch 0 Loss 0.2012
Epoch 7 Batch 100 Loss 0.1468
Epoch 7 Batch 200 Loss 0.2198
Epoch 7 Batch 300 Loss 0.2109
Epoch 7 Loss 0.2155
Time taken for 1 epoch 16.99945044517517 sec

Epoch 8 Batch 0 Loss 0.1343
Epoch 8 Batch 100 Loss 0.1683
Epoch 8 Batch 200 Loss 0.1547
Epoch 8 Batch 300 Loss 0.1345
Epoch 8 Loss 0.1589
Time taken for 1 epoch 17.30172872543335 sec

Epoch 9 Batch 0 Loss 0.1193
Epoch 9 Batch 100 Loss 0.1181
Epoch 9 Batch 200 Loss 0.1104
Epoch 9 Batch 300 Loss 0.1278
Epoch 9 Loss 0.1278
Time taken for 1 epoch 17.062464952468872 sec

Epoch 10 Batch 0 Loss 0.0915
Epoch 10 Batch 100 Loss 0.0890
Epoch 10 Batch 200 Loss 0.1234
Epoch 10 Batch 300 Loss 0.1449
Epoch 10 Loss 0.1016
Time taken for 1 epoch 17.3432514667511 sec

অনুবাদ করা

• মূল্যায়ন ফাংশনটি প্রশিক্ষণের লুপের সমান, আমরা এখানে শিক্ষক জোর করে ব্যবহার করি না except প্রতিটি সময় পদক্ষেপে ডিকোডারের ইনপুট হ'ল লুকানো অবস্থা এবং এনকোডার আউটপুট সহ পূর্ববর্তী ভবিষ্যদ্বাণী।
• যখন মডেলটি শেষ টোকেনটির পূর্বাভাস দেয় তখন ভবিষ্যদ্বাণী করা বন্ধ করুন ।
• এবং প্রতিবার পদক্ষেপের জন্য মনোযোগের ওজন সংরক্ষণ করুন ।

def evaluate(sentence):
    attention_plot = np.zeros((max_length_targ, max_length_inp))

    sentence = preprocess_sentence(sentence)

    inputs = [inp_lang.word_index[i] for i in sentence.split(' ')]
    inputs = tf.keras.preprocessing.sequence.pad_sequences([inputs],
                                                           maxlen=max_length_inp,
                                                           padding='post')
    inputs = tf.convert_to_tensor(inputs)

    result = ''

    hidden = [tf.zeros((1, units))]
    enc_out, enc_hidden = encoder(inputs, hidden)

    dec_hidden = enc_hidden
    dec_input = tf.expand_dims([targ_lang.word_index['<start>']], 0)

    for t in range(max_length_targ):
        predictions, dec_hidden, attention_weights = decoder(dec_input,
                                                             dec_hidden,
                                                             enc_out)

        # storing the attention weights to plot later on
        attention_weights = tf.reshape(attention_weights, (-1, ))
        attention_plot[t] = attention_weights.numpy()

        predicted_id = tf.argmax(predictions[0]).numpy()

        result += targ_lang.index_word[predicted_id] + ' '

        if targ_lang.index_word[predicted_id] == '<end>':
            return result, sentence, attention_plot

        # the predicted ID is fed back into the model
        dec_input = tf.expand_dims([predicted_id], 0)

    return result, sentence, attention_plot

# function for plotting the attention weights
def plot_attention(attention, sentence, predicted_sentence):
    fig = plt.figure(figsize=(10,10))
    ax = fig.add_subplot(1, 1, 1)
    ax.matshow(attention, cmap='viridis')

    fontdict = {'fontsize': 14}

    ax.set_xticklabels([''] + sentence, fontdict=fontdict, rotation=90)
    ax.set_yticklabels([''] + predicted_sentence, fontdict=fontdict)

    ax.xaxis.set_major_locator(ticker.MultipleLocator(1))
    ax.yaxis.set_major_locator(ticker.MultipleLocator(1))

    plt.show()

def translate(sentence):
    result, sentence, attention_plot = evaluate(sentence)

    print('Input: %s' % (sentence))
    print('Predicted translation: {}'.format(result))

    attention_plot = attention_plot[:len(result.split(' ')), :len(sentence.split(' '))]
    plot_attention(attention_plot, sentence.split(' '), result.split(' '))

সর্বশেষতম চেকপয়েন্ট এবং পরীক্ষা পুনরুদ্ধার করুন

# restoring the latest checkpoint in checkpoint_dir
checkpoint.restore(tf.train.latest_checkpoint(checkpoint_dir))

<tensorflow.python.training.tracking.util.CheckpointLoadStatus at 0x7f687502d9e8>

translate(u'hace mucho frio aqui.')

Input: <start> trata de averiguarlo . <end>
Predicted translation: try to figure it out . <end> 

translate(u'esta es mi vida.')

Input: <start> esta es mi vida . <end>
Predicted translation: this is my life . <end> 

translate(u'¿todavia estan en casa?')

Input: <start> ¿ todavia estan en casa ? <end>
Predicted translation: are we still at home now ? <end>

# wrong translation
translate(u'trata de averiguarlo.')

Input: <start> trata de averiguarlo . <end>
Predicted translation: try to figure it out . <end>