Search⌘ K
AI Features
Home
Courses
Building Advanced Deep Learning and NLP Projects

Project Creation: Part Three

Explore how to create a character-level recurrent neural network model for text decryption. Learn to define model architecture with GRU layers, preprocess and reshape input data, train the model effectively, and convert numeric outputs back to readable text. Gain practical skills to apply RNNs in decoding encrypted text accurately.

We'll cover the following...

In the previous lesson, we discussed a lot of topics. Now we will be moving ahead and building our model architecture.

Model architecture

We will define a function that will accept four parameters: the input shape, the output sequence length, the vocabulary size of encrypted text, and the vocabulary size of the plaintext.

Python 3.5
def simple_model(input_shape, output_sequence_length, code_vocab_size, plaintext_vocab_size):
    
    learning_rate = 1e-3
    input_seq = Input(input_shape[1:])
    rnn = GRU(64, return_sequences=True)(input_seq)
    logits = TimeDistributed(Dense(plaintext_vocab_size))(rnn)
    model = Model(input_seq, Activation('softmax')(logits))
    model.compile(loss=sparse_categorical_crossentropy,
                  optimizer=Adam(learning_rate),
                  metrics=['accuracy'])
    return model
print("Function Created Successfully!")

Explanation:

  • We first defined our function, which accepts the four parameters which we discussed above.
  • On line 3, we defined our learning rate for the network.
  • On line 5, we created an Input() object by specifying the shape of the input data as a parameter.
  • On line 6, we defined our GRU layer with 64 units. This means that RNN will be unrolled 64 times. Also, the return_sequences=True specifies whether to return the last output in the output sequence or not.
  • On line 7, we defined our TimeDistributed layer by passing a dense layer as its parameter.
  • We also specified input_seq and rnn after the function
...