kenhenry / tensorflow-seq2seq-chatbot Goto Github PK

what is exactly happening during testing after the user input get tokenized (i.e)what we get at encoder_input,decoder_input) and how the data get fetch based on given input.I am in need so please explain it briefly

During training the seq2seq chatbot ,why it take long time to train even if I give small amount of data .what is the purpose of test.enc and test.dec

Here is the detial error message :
File "execute.py", line 218, in decode
enc_vocab, _ = data_utils.initialize_vocabulary(enc_vocab_path)
File "E:\tensorflow-seq2seq-chatbot-master\tensorflow-seq2seq-chatbot-master\data_utils.py", line 89, in initialize_vocabulary
rev_vocab.extend(f.readlines())
File "C:\Program Files\Python35\lib\site-packages\tensorflow\python\lib\io\file_io.py", line 131, in readlines
s = self.readline()
File "C:\Program Files\Python35\lib\site-packages\tensorflow\python\lib\io\file_io.py", line 124, in readline
return compat.as_str_any(self._read_buf.ReadLineAsString())
File "C:\Program Files\Python35\lib\site-packages\tensorflow\python\util\compat.py", line 106, in as_str_any
return as_str(value)
File "C:\Program Files\Python35\lib\site-packages\tensorflow\python\util\compat.py", line 84, in as_text
return bytes_or_text.decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x92 in position 1: invalid start byte

Hi guys. I am new to tensorflow and just trying out couples of examples.
I am using Tensorflow 1.0.1 and python 2.7 w anaconda.

I think it's because of the function relocations, there are couples of issue happens, when I am running the train section.
.
.
.

ValueError: Shape must be rank 2 but is rank 1 for 'model_with_buckets/sequence_loss/sequence_loss_by_example/sampled_softmax_loss/LogUniformCandidateSampler' (op: 'LogUniformCandidateSampler') with input shapes: [?].

Not sure the should I change the input or what, anyone have some clues?

Thanks a lot.

===================================================
Update #1
Turn out I have sort this one out , but I encounter another one, seems something wrong with model_with_buckets()
Here I list all the change I made.

1. In seq2seq2.py
   i) from tensorflow.models.rnn.translate import data_utils change -> import data_utils
   ii) nn.rnn_cell change -> contrib.rnn
   iii) nn.seq2seq change -> contrib.legacy_seq2seq
   iv) fixing ValueError , interchange "label" and "input"

In tf 0.12.0: Definition : tf.nn.sampled_softmax_loss(weights, biases, inputs, labels, num_sampled, num_classes, num_true=1, sampled_values=None, remove_accidental_hits=True, partition_strategy="mod", name="sampled_softmax_loss")

In tf 1.0: tf.nn.sampled_softmax_loss(weights, biases, labels, inputs, num_sampled, num_classes, num_true=1, sampled_values=None, remove_accidental_hits=True, partition_strategy='mod', name='sampled_softmax_loss')

2. For missing train.enc error, check data/ exists or not

New Error:

File "/Users/ATC/Desktop/TensorFlow/6. Chatbot/seq2seq_model.py", line 149, in init
softmax_loss_function=softmax_loss_function)
TypeError: Input 'y' of 'Mul' Op has type float32 that does not match type int32 of argument 'x'.

Hi kenhenry,

thanks for sharing code. i have tried with ubuntu chat log. but i'm facing some common issues like generic responses issues and inconsistent reply.

is this model ( code shared by you ) handling some common issues related to seq2seq model on chatbot like replying generic responses (ex: 'Okay','No','Yes'..) and inconsistent replies for paraphrased contexts but with the same sense.
like
Q-where do you live now?
A-I live in Chicago.
Q-In which country do you live now?
A-Iceland, you?
...like this. inconsistent reply.

these are the most common issues in this model. are you handling this ? if not what's your thought how can we handle these issues?

definitely this is not issues related to code. just don't have other option to get answer for above questions from you.

Thanks,
muna

Firstly thank you for the updated execute.py, I couldn't for the life of me get Suriya's script working but given Im very very new ubuntu, python and TF I though it was my fault, your git worked first time.

having said that this is probably such a stupid question but .... when does this stop (or how do I stop it cleanly)

I thought it might stop

"if counter % 100000 == 0:" but

global step 101000 learning rate 0.4432 step-time 0.63 perplexity 2.78
eval: bucket 0 perplexity 130.77
eval: bucket 1 perplexity 161.28
eval: bucket 2 perplexity 91.37
eval: bucket 3 perplexity 185.74

...... maybe don't worry about the cleanly ... turns out trying to copy from a putty window sends a keyboard interrupt.

:(

Thanks again for your work on this.

J

While running execute.py with mode=train, program gets stuck at reading data 18135. Anyone encountered the same issue? I have come across this issue on multiple codes of seq2seq models

I will like run in GPU mode