• Search by Module
• Search by Words
• Search Projects

• Java
• Python
• JavaScript
• TypeScript
• C++
• Scala
• Blog

Project: lang2program (GitHub Link)

• lang2program-master
  • dependency
    • data_directory.py
    • __init__.pyc
    • __init__.py
    • data_directory.pyc
  • gtd
    • log.py
    • profile_imports.py
    • postgres.py
    • ml
      • framework.py
      • profile.py
      • seq_batch.py
      • model.py
      • experiment.py
      • __init__.py
      • utils.py
      • vocab.py
    • graph.py
    • text.py
    • lm.py
    • plot.py
    • git_utils.py
    • persist.py
    • codalab.py
    • io.py
    • chrono.py
    • __init__.py
    • utils.py
    • tests
      • ml
        • test_seq_batch.py
        • test_utils.py
        • test_vocab.py
        • test_model.py
        • test_framework.py
        • __init__.py
      • test_utils.py
      • test_persist.py
      • test_log.py
      • test_io.py
      • test_lm.py
      • test_graph.py
      • __init__.py
  • strongsup
    • visualizer.py
    • path_checker.py
    • case_weighter.py
    • decoder.py
    • predicate.py
    • results
      • recipe.py
      • table_drawer.py
      • result_value.py
      • entry.py
      • __init__.py
      • tracker.py
      • entry_selector.py
    • example_factory.py
    • exploration_policy.py
    • static_exploration.py
    • evaluation.py
    • world.py
    • tables
      • path_checker.py
      • graph.py
      • predicate.py
      • example_factory.py
      • structure.py
      • world.py
      • domain.py
      • executor.py
      • predicates_computer.py
      • value.py
      • __init__.py
      • utils.py
    • rlong
      • state.py
      • path_checker.py
      • predicate.py
      • example_factory.py
      • exploration_policy.py
      • world.py
      • domain.py
      • executor.py
      • predicates_computer.py
      • value.py
      • __init__.py
    • domain.py
    • parse_case.py
    • executor.py
    • predicates_computer.py
    • value_function.py
    • value.py
    • example.py
    • experiment.py
    • embeddings.py
    • __init__.py
    • parse_model.py
    • utils.py
    • tests
      • test_utils.py
      • test_decoder.py
      • results
        • tensorboard
          • events.out.tfevents.1487580318.jagupard15.stanford.edu
        • other_tensorboard
        • test_tracker.py
      • __pycache__
        • test_parse_case.cpython-27-PYTEST.pyc
        • test_parse_model.cpython-27-PYTEST.pyc
        • test_decoder.cpython-27-PYTEST.pyc
        • test_experiment.cpython-27-PYTEST.pyc
      • test_experiment.py
      • tables
        • test_utils.py
        • __pycache__
          • test_graph.cpython-27-PYTEST.pyc
          • test_utils.cpython-27-PYTEST.pyc
          • test_predicates_computer.cpython-27-PYTEST.pyc
          • test_structure.cpython-27-PYTEST.pyc
          • test_executor.cpython-27-PYTEST.pyc
        • test_predicates_computer.py
        • test_graph.py
        • test_structure.py
        • test_executor.py
      • rlong
        • test_executor.py
        • test_exploration_policy.py
      • test_parse_model.py
      • test_value_function.py
      • test_example.py
      • __init__.py
      • utils.py
      • test_parse_case.py
    • dataset.py
  • LICENSE
  • third-party
    • gtd
      • gtd
        • log.py
        • profile_imports.py
        • postgres.py
        • ml
          • framework.py
          • profile.py
          • seq_batch.py
          • model.py
          • experiment.py
          • __init__.py
          • utils.py
          • vocab.py
        • graph.py
        • text.py
        • lm.py
        • plot.py
        • git_utils.py
        • persist.py
        • codalab.py
        • io.py
        • chrono.py
        • __init__.py
        • utils.py
        • tests
          • ml
            • test_seq_batch.py
            • test_utils.py
            • test_vocab.py
            • test_model.py
            • test_framework.py
            • __init__.py
          • test_utils.py
          • test_persist.py
          • test_log.py
          • test_io.py
          • test_lm.py
          • test_graph.py
          • __init__.py
      • setup.py
      • README.md
      • scripts
        • git_logs.py
      • requirements.txt
      • .gitignore
  • configs
    • rlong
      • best-scene.txt
      • best-tangrams.txt
      • default-base.txt
      • debug-base.txt
      • dataset-mixins
        • alchemy.txt
        • tangrams.txt
        • scene.txt
      • config-mixins
        • baseline=0.01.txt
        • alpha=0.txt
        • beta=0.txt
        • multi-step-train.txt
        • only-use-stack-emb.txt
        • epsilon=0.05.txt
        • beta=0.25.txt
        • beam-search.txt
        • epsilon=0.3.txt
        • tangrams-reinforce-natural-death.txt
        • baseline=0.00001.txt
        • alchemy-natural-death.txt
        • epsilon=0.1.txt
        • batch-size=64.txt
        • beta=0.5.txt
        • stale-beam-age=3.txt
        • alpha=1.2.txt
        • tangrams-natural-death.txt
        • batched-reinforce-lookahead.txt
        • test_beam_size=256.txt
        • cpu.txt
        • stack-emb.txt
        • beta=1.txt
        • use-stack-emb.txt
        • batch-size=256.txt
        • beta=2.txt
        • baseline=0.003.txt
        • adam-rate=0.3.txt
        • baseline=0.03.txt
        • batched-reinforce-gamma=10.txt
        • adam-rate=0.01.txt
        • test_beam_size=1.txt
        • batched-reinforce-uniform.txt
        • alchemy-reinforce-natural-death.txt
        • batched-reinforce-gamma=0.3.txt
        • adam-rate=0.003.txt
        • adam-rate=0.1.txt
        • stale-beam-age=5.txt
        • epsilon=0.12.txt
        • sgd.txt
        • stale-beam-age=10.txt
        • scene-reinforce-natural-death.txt
        • epsilon=0.02.txt
        • batched-reinforce-basic.txt
        • alpha=0.5.txt
        • batched-reinforce-epsilon=0.2.txt
        • all-substring-train.txt
        • baseline=0.1.txt
        • scene-natural-death.txt
        • train_beam_size=128.txt
        • epsilon=0.2.txt
        • beta=0.75.txt
        • epsilon=0.25.txt
        • no-save.txt
        • train_beam_size=1.txt
        • train_beam_size=256.txt
        • baseline=0.0001.txt
        • batched-reinforce-zombie.txt
        • baseline=0.001.txt
        • stale-beam-age=20.txt
        • adam-rate=0.03.txt
        • indep-utt-expl.txt
        • alpha=0.8.txt
        • batched-reinforce-gamma=0.03.txt
        • epsilon=0.15.txt
        • epsilon=0.08.txt
        • batched-reinforce-epsilon=0.1.txt
        • batched-reinforce-penalty.txt
      • best-alchemy.txt
  • README.md
  • launch_docker
  • scripts
    • result.py
    • main.py
  • requirements.txt
  • scone.md
  • Dockerfile
  • .gitignore

Introduction

Authors: Kelvin Guu, Panupong (Ice) Pasupat, Evan Zheran Liu, Percy Liang

Source code accompanying our ACL 2017 paper, From Language to Programs: Bridging Reinforcement Learning and Maximum Marginal Likelihood.

Also see:

• An introduction to SCONE, the context-dependent semantic parsing dataset that we evaluate on.
• Reproducible experiments on our worksheet at CodaLab.org.

Setup

First, download the repository and necessary data.

$ git clone https://github.com/kelvinguu/lang2program.git
$ mkdir -p lang2program/data
$ cd lang2program/data
$ wget http://nlp.stanford.edu/data/glove.6B.zip  # GloVe vectors
$ unzip glove.6B.zip -d glove.6B
$ wget https://nlp.stanford.edu/projects/scone/scone.zip  # SCONE dataset
$ unzip scone.zip

The resulting data directory should look like this:

• data/
  • glove.6B/
  • rlong/

Now, start the project's Docker container (you will need to install Docker). The container has all the required software dependencies installed.

$ cd ..
$ ./launch_docker

This script will download the appropriate Docker image if it is not already on your machine. Downloading the image may take a while.

Inside the container, your Git repository will be mounted at /lang2program. All subsequent instructions in this README should be performed inside the container. To exit the container, type exit, just as you would exit bash.

Training a model

To launch a new training run:

$ cd /lang2program
$ python scripts/main.py configs/rlong/best-scene.txt

To run a different configuration, replace configs/rlong/best-scene.txt with your own config file. See configs/rlong/default-base.txt and configs/rlong/dataset-mixins/scene.txt for reasonable starting points. These files are in HOCON syntax.

On stdout, the script will print out the experiment's ID number.

Data for this experiment will be saved to the directory /lang2program/data/experiments/<experiment_id>, containing the following files:

• config.txt
  • The config file for this training run
• checkpoints/
  • TensorFlow checkpoints, saved during training
• tensorboard/
  • TensorBoard log files
• codalab.json
  • The results of periodic evaluation are saved here.

Program syntax

See scone.md for more information.