Please make sure that this is a Bug or a Feature Request and provide all applicable information asked by the template. If your issue is an implementation question, please ask your question on StackOverflow or on the Keras Slack channel instead of opening a GitHub issue.

System information

Have I written custom code (as opposed to using example directory):
OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
TensorFlow backend (yes / no): Yes
TensorFlow version: 1.13.1
Keras version: 2.2.4
Python version: 3.6.8
CUDA/cuDNN version: No CUDA
GPU model and memory: No GPU

You can obtain the TensorFlow version with:
python -c "import tensorflow as tf; print(tf.GIT_VERSION, tf.VERSION)"
You can obtain the Keras version with:
python -c 'import keras as k; print(k.version)'

Describe the current behavior

I'm prototyping as Seq2Seq Model as described in the https://blog.keras.io/a-ten-minute-introduction-to-sequence-to-sequence-learning-in-keras.html blog post. It works fine with a LSTM layer for inference, using GRU for training works OK, but the code for inference doesn't work at all, I suspect that my code must have some error or something like like.

Describe the expected behavior

I haven't code for inference in the blog post for GRUs, does anyone have used before?

Code to reproduce the issue
Provide a reproducible test case that is the bare minimum necessary to generate the problem.

Here's the code: from keras.models import Model from keras.layers import Input, Dense, GRU, Bidirectional import numpy as np

input_texts = [] target_texts = [] chars = set() num_sentences = 1024

for i, line in enumerate(open('fra.txt', 'r', encoding='utf-8')): line = line.lower() line = line.replace('\n', '') inp, out = line.split('\t') input_texts.append(inp) out = '\t' + out + '\n' target_texts.append(out)

for ch in inp:
    if ch not in chars:
        chars.add(ch)

for ch in out:
    if ch not in chars:
        chars.add(ch)

if i > num_sentences:
    break

chr2idx = {} idx2chr = {}

for i, ch in enumerate(chars): idx2chr[i] = ch chr2idx[ch] = i

print(chr2idx)

max_seq = 64 num_tokens = len(chars) num_samples = len(input_texts)

encoder_input_data = np.zeros((num_samples, max_seq, num_tokens), dtype='float32') decoder_input_data = np.zeros((num_samples, max_seq, num_tokens), dtype='float32') decoder_target_data = np.zeros((num_samples, max_seq, num_tokens), dtype='float32')

for i, (input_text, target_text) in enumerate(zip(input_texts, target_texts)): for t, char in enumerate(input_text): encoder_input_data[i, t, chr2idx[char]] = 1.0 for t, char in enumerate(target_text): decoder_input_data[i, t, chr2idx[char]] = 1.0 if t > 0: decoder_target_data[i, t - 1, chr2idx[char]] = 1.0

def Encoder(latent_dim, num_tokens): _input = Input(shape=(None, num_tokens)) encoder = Bidirectional(GRU(latent_dim, return_state=True)) outputs, hf, hb = encoder(_input) return {'input' : _input, 'encoder' : encoder, 'states' : [hf, hb]}

def Decoder(latent_dim, num_tokens, states): _input = Input(shape=(None, num_tokens)) decoder = Bidirectional(GRU(latent_dim, return_sequences=True)) outputs = decoder(_input, initial_state=states) fc = Dense(num_tokens, activation='softmax') output = fc(outputs) return {'input' : _input, 'decoder' : decoder, 'output' : output}

latent_dim = 128

encoder = Encoder(latent_dim, num_tokens) decoder = Decoder(latent_dim, num_tokens, encoder['states'])

model = Model([encoder['input'], decoder['input']], decoder['output']) model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['acc']) model.summary()

encoder_infer = Model(encoder['input'], encoder['states']) h = Input(shape=(None, num_tokens)) outputs, hf, hb = decoder'decoder'

output = decoder'output' decoder_infer = Model([decoder['input']], [decoder['output']] + [hf, hb])

cont = encoder_infer.predict(encoder_input_data[0]) out = decoder_infer.predict(cont) print(out)

model.fit([encoder_input_data, decoder_input_data], decoder_target_data, batch_size=128, epochs=1000)

Other info / logs
Include any logs or source code that would be helpful to diagnose the problem. If including tracebacks, please include the full traceback. Large logs and files should be attached.

This is the error message that I getting: Traceback (most recent call last): File "bi_gru.py", line 84, in outputs, hf, hb = decoder'decoder' File "C:\Program Files\Python36\lib\site-packages\tensorflow\python\framework\ops.py", line 442, in iter "Tensor objects are only iterable when eager execution is " TypeError: Tensor objects are only iterable when eager execution is enabled. To iterate over this tensor use tf.map_fn.

Cheers.

keras-team / keras

Seq2Seq Model with GRU tf.map_fn error #12825

model.fit([encoder_input_data, decoder_input_data], decoder_target_data, batch_size=128, epochs=1000)