web-dev-qa-db-fra.com

Keras seq2seq - Incorporation de mots

Je travaille sur un chatbot génératif basé sur seq2seq à Keras. J'ai utilisé le code de ce site: https://machinelearningmastery.com/develop-encoder-decoder-model-sequence-sequence-prediction-keras/

Mes modèles ressemblent à ceci:

# define training encoder
encoder_inputs = Input(shape=(None, n_input))
encoder = LSTM(n_units, return_state=True)
encoder_outputs, state_h, state_c = encoder(encoder_inputs)
encoder_states = [state_h, state_c]

# define training decoder
decoder_inputs = Input(shape=(None, n_output))
decoder_lstm = LSTM(n_units, return_sequences=True, return_state=True)
decoder_outputs, _, _ = decoder_lstm(decoder_inputs, initial_state=encoder_states)
decoder_dense = Dense(n_output, activation='softmax')
decoder_outputs = decoder_dense(decoder_outputs)
model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

# define inference encoder
encoder_model = Model(encoder_inputs, encoder_states)

# define inference decoder
decoder_state_input_h = Input(shape=(n_units,))
decoder_state_input_c = Input(shape=(n_units,))
decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]
decoder_outputs, state_h, state_c = decoder_lstm(decoder_inputs, initial_state=decoder_states_inputs)
decoder_states = [state_h, state_c]
decoder_outputs = decoder_dense(decoder_outputs)
decoder_model = Model([decoder_inputs] + decoder_states_inputs [decoder_outputs] + decoder_states)

Ce réseau de neurones est conçu pour fonctionner avec un seul vecteur codé à chaud, et l’entrée sur ce réseau ressemble par exemple à ceci:

[[[0. 0. 0. 0. 1. 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. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0. 0. 0.]
  [0. 0. 0. 0. 0. 0. 0. 1. 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. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0. 0. 0.]
  [0. 0. 1. 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. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0. 0. 0.]]
  [[0. 0. 0. 0. 1. 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. 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. 0. 0. 0. 1. 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.]
  [0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 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.]]]

Comment puis-je reconstruire ces modèles pour qu'ils fonctionnent avec des mots? J'aimerais utiliser le calque incorporant Word, mais je ne sais pas comment connecter ce calque à ces modèles. 

Mon entrée devrait être [[1,5,6,7,4], [4,5,7,5,4], [7,5,4,2,1]] où int nombres sont des représentations de mots.

J'ai tout essayé mais je continue à avoir des erreurs. Pouvez-vous m'aider s'il vous plaît? 

pythonneural-networkkeraschatbotembedding
5
Lukáš Richtarik

Je l'ai finalement fait. Voici le code: 

Shared_Embedding = Embedding(output_dim=embedding, input_dim=vocab_size, name="Embedding")

encoder_inputs = Input(shape=(sentenceLength,), name="Encoder_input")
encoder = LSTM(n_units, return_state=True, name='Encoder_lstm') 
Word_embedding_context = Shared_Embedding(encoder_inputs) 
encoder_outputs, state_h, state_c = encoder(Word_embedding_context) 
encoder_states = [state_h, state_c] 
decoder_lstm = LSTM(n_units, return_sequences=True, return_state=True, name="Decoder_lstm")

decoder_inputs = Input(shape=(sentenceLength,), name="Decoder_input")
Word_embedding_answer = Shared_Embedding(decoder_inputs) 
decoder_outputs, _, _ = decoder_lstm(Word_embedding_answer, initial_state=encoder_states) 
decoder_dense = Dense(vocab_size, activation='softmax', name="Dense_layer") 
decoder_outputs = decoder_dense(decoder_outputs) 

model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

encoder_model = Model(encoder_inputs, encoder_states) 

decoder_state_input_h = Input(shape=(n_units,), name="H_state_input") 
decoder_state_input_c = Input(shape=(n_units,), name="C_state_input") 
decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c] 
decoder_outputs, state_h, state_c = decoder_lstm(Word_embedding_answer, initial_state=decoder_states_inputs) 
decoder_states = [state_h, state_c] 
decoder_outputs = decoder_dense(decoder_outputs)

decoder_model = Model([decoder_inputs] + decoder_states_inputs, [decoder_outputs] + decoder_states)

"modèle" correspond au modèle d'apprentissage encoder_model et decoder_model sont des modèles d'inférence

4
Lukáš Richtarik

Ci-dessous, dans la section FAQ de cet exemple, ils fournissent un exemple d'utilisation des opérations d'immersion avec seq2seq. Je suis en train de déterminer moi-même l'étape d'inférence. Je posterai ici quand je l’aurai. https://blog.keras.io/a-ten-minute-introduction-to-sequence-to-sequence-learning-in-keras.html

0
emericw