crazydonkey200 / neural-symbolic-machines Goto Github PK

Neural Symbolic Machines is a framework to integrate neural networks and symbolic representations using reinforcement learning, with applications in program synthesis and semantic parsing.

License: Apache License 2.0

Shell 4.05% Python 93.22% Jupyter Notebook 2.73%

deep-learning deep-reinforcement-learning tensorflow program-synthesis semantic-parsing rnn reinforcement-learning

neural-symbolic-machines's People

Contributors

Stargazers

Watchers

Forkers

daming-lu hankersia arthesfan chenglongchen michael-wzhu jreuben11 agarwl kesyren murtyshikhar songgc rulai-yananxie fbfra baylee001 yucoian anirband qiuwei codeaudit satpreetsingh kastnerkyle lyztyj jinyeong henrywoodotc zhihaolzh tk1363704 pcyin dungtn rshigemura 0101011 ckeshava ameyagodbole knightmarehs arisosoftware gokasiko daominglyu hyzcn carlito97 seanhtchoi freesia-sherry stuartchan xrosliang barryzm runningstallion singingtower huanpass candyyinzx kuonanhong tmr adapptgru scape1989 wxwoods odellus ayoob7 nnwhisperer alan-qiu mtoub mistobaan xiangnanxing tedrepo 3i-hust-ps yuanbei sandy4321 alexanderdurr qishi21 mrrobot2211 ajunlonglive graphgrailai skstcode42 iq-scm zebrajack

neural-symbolic-machines's Issues

training stop

I have edit the whole code to python3.

And the program stop here:

ctrl + c shows

  File "experiment_mapo.py", line 1274, in <module>
    tf.app.run()
  File "/data4/tong.guo/Py36/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 125, in run
    _sys.exit(main(argv))
  File "experiment_mapo.py", line 1270, in main
    run_experiment()
  File "experiment_mapo.py", line 704, in run_experiment
    actor.join()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/process.py", line 124, in join
    res = self._popen.wait(timeout)
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/popen_fork.py", line 50, in wait
    return self.poll(os.WNOHANG if timeout == 0.0 else 0)
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/popen_fork.py", line 28, in poll
    pid, sts = os.waitpid(self.pid, flag)

and

Process actor_15:
Process actor_3:
Process actor_14:
Traceback (most recent call last):
Process actor_6:
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/popen_fork.py", line 28, in poll
    pid, sts = os.waitpid(self.pid, flag)
KeyboardInterrupt
Traceback (most recent call last):
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/process.py", line 258, in _bootstrap
    self.run()
  File "experiment_mapo.py", line 962, in run
    new_ckpt = self.ckpt_queue.get()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/queues.py", line 93, in get
    with self._rlock:
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/synchronize.py", line 96, in __enter__
    return self._semlock.__enter__()
KeyboardInterrupt

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
Traceback (most recent call last):
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/process.py", line 261, in _bootstrap
    util._exit_function()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/util.py", line 322, in _exit_function
    _run_finalizers()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/util.py", line 262, in _run_finalizers
    finalizer()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/util.py", line 186, in __call__
    res = self._callback(*self._args, **self._kwargs)
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/queues.py", line 191, in _finalize_join
    thread.join()
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/process.py", line 258, in _bootstrap
    self.run()
  File "/data4/tong.guo/Py36/lib/python3.6/threading.py", line 1056, in join
    self._wait_for_tstate_lock()
  File "experiment_mapo.py", line 962, in run
    new_ckpt = self.ckpt_queue.get()
  File "/data4/tong.guo/Py36/lib/python3.6/threading.py", line 1072, in _wait_for_tstate_lock
    elif lock.acquire(block, timeout):
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/queues.py", line 93, in get
    with self._rlock:
  File "/data4/tong.guo/Py36/lib/python3.6/multiprocessing/synchronize.py", line 96, in __enter__
    return self._semlock.__enter__()
KeyboardInterrupt

For Wikisql, different agg may lead to same result. How do you deal with this problem?

Thank you!

Unable to install dependencies

Can you please specify the configuration of the emvironment? I am using Anaconda2(because python2 is required) package manager. All the libraries in requirements.txt have been successfully installed.

Specifically, I'm running into this error:

Traceback (most recent call last):
  File "../experiment.py", line 29, in <module>
    from nsm import word_embeddings
  File "/home/16co108/mapo/projects/neural-symbolic-machines/table/wikisql/nsm/word_embeddings.py", line 3, in <module>
    import gensim
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/__init__.py", line 6, in <module>
    from gensim import parsing, matutils, interfaces, corpora, models, similarities, summarization, utils  # noqa:F401
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/parsing/__init__.py", line 4, in <module>
    from .preprocessing import (remove_stopwords, strip_punctuation, strip_punctuation2,  # noqa:F401
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/parsing/preprocessing.py", line 40, in <module>
    from gensim import utils
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/utils.py", line 41, in <module>
    import scipy.sparse
  File "/opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/__init__.py", line 229, in <module>
    from .csr import *
  File "/opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/csr.py", line 15, in <module>
    from ._sparsetools import csr_tocsc, csr_tobsr, csr_count_blocks, \
ImportError: /usr/lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by /opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/_sparsetools.so)
Traceback (most recent call last):
  File "../experiment.py", line 29, in <module>
    from nsm import word_embeddings
  File "/home/16co108/mapo/projects/neural-symbolic-machines/table/wikisql/nsm/word_embeddings.py", line 3, in <module>
    import gensim
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/__init__.py", line 6, in <module>
    from gensim import parsing, matutils, interfaces, corpora, models, similarities, summarization, utils  # noqa:F401
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/parsing/__init__.py", line 4, in <module>
    from .preprocessing import (remove_stopwords, strip_punctuation, strip_punctuation2,  # noqa:F401
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/parsing/preprocessing.py", line 40, in <module>
    from gensim import utils
  File "/home/16co108/.local/lib/python2.7/site-packages/gensim/utils.py", line 41, in <module>
    import scipy.sparse
  File "/opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/__init__.py", line 229, in <module>
    from .csr import *
  File "/opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/csr.py", line 15, in <module>
    from ._sparsetools import csr_tocsc, csr_tobsr, csr_count_blocks, \
ImportError: /usr/lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by /opt/ohpc/pub/anaconda2/5.0.1/lib/python2.7/site-packages/scipy/sparse/_sparsetools.so)

Coverage of the domain specific language

Hi, Chen:

Can you give the coverage of your domain specific language for wikitable and wikisql? I am trying to reproduce this language and my system can cover around 74% questions in training set for wikitable. I want to make sure if it matches your number.

Thanks!

What happens in test time?

NSM only use question-answer pair to train.
And in test time, NSM generate the SQL logic form.
Am I right?

Different vocab for different en_vocab files.

neural-symbolic-machines/nsm/word_embeddings.py

Lines 14 to 16 in 231bdc3

    
           with open(vocab_file, 'r') as f: 
        
             tks = json.load(f) 
        
           self.vocab = dict(zip(tks, range(len(tks))))

Due to the above line, the vocab in embedding model i.e. embedding_model.vocab is different (i.e mapping from token to index would be different) depending on the en_vocab_count_i.json file you use (I tried it for i = 1, 5).

This doesn't seem right because the embeddings loaded for a given token would be dependent on the vocabulary you used.

neural-symbolic-machines/nsm/word_embeddings.py

Lines 21 to 24 in 231bdc3

    
           def __getitem__(self, word): 
        
             if word in self.vocab: 
        
               index = self.vocab[word] 
        
               return self.embedding_mat[index]

Imported modules not defined

import wtq_evaluator
import wtq_utils

In the file preprocess.py, the above modules are imported but not defined. I guess the correct imports should be:

import evaluator as wtq_evaluator
import ..utils as wtq_utils

Hard to find the code where the reward guide the training.

Sorry to trouble you, but there are too much code.
Would you please point out the code place about the core of REINFORCE algorithm.
Thank you very much.

The paper says it also need a number of labeled SQL to train?

Thank you very much.

Is sequence entropy defined only each action in a sequence?

neural-symbolic-machines/nsm/graph_factory.py

Line 347 in 231bdc3

seq_entropy = (

Should there be another reduce_sum over sequence length dimension for sequence entropy term calculation?

NSM also use database table header to train, am I right?

In Wikisql.
Not only question-answer pair.
Thank you!

`scale_rewards` function uses undefined variables

def scale_rewards(samples, scale):
  """Weight each samples with the weight. Reflected on rewards."""
  scaled_rewards = [list(scale * np.array(s.rewards)) for s in samples]
  new_samples = []
  for s, p in zip(samples, scaled_probs):
    new_samples.append(
        sample._replace(traj=traj._replace(rewards=scaled_rewards)))
  return new_samples

In the function in agent_factory.py, the variables sample and scaled_probs are undefined.

请问下，除了warm start，还有哪些设计大幅提升了结果？

reward方面，predict answer和label answer一致就是好的reward，还有其他reward相关设计提升结果吗？
感谢！
@crazydonkey200

The experiment on WebquestionSP

Could you please also publish the experiment on WebquestionSP?

Why targets=actions?

In https://github.com/crazydonkey200/neural-symbolic-machines/blob/master/nsm/agent_factory.py

In https://github.com/crazydonkey200/neural-symbolic-machines/blob/master/nsm/model_factory.py

I guess targets should at least be related to rewards.
Thank you!!

Where does load_programs 's data come from?

in experiment.py
Thank you!
@crazydonkey200

What is the ValueEstimator of NSM?

I read that REINFORCE algorithm have a ValueApproximator and a PolicyEstimator.
Thank you!

reproducing results paper

Hi!,
I was playing with your code, great work! I am trying to reproduce the results from your paper on WikiSQL. However, when using run.sh I get results in the 70.3 ballpark (on dev set) instead of the reported 72.2%. Are there any parameters I need to change to get the reported results?

Thanks in advance!

attn_inputs = en_seq_outputs

	with open(vocab_file, 'r') as f:
	tks = json.load(f)
	self.vocab = dict(zip(tks, range(len(tks))))

	def __getitem__(self, word):
	if word in self.vocab:
	index = self.vocab[word]
	return self.embedding_mat[index]