Implementation of the paper "Generating Symbolic Reasoning Problems with Transformer GANs"

Related tags

Deep LearningTGAN-SR
Overview

Generating Symbolic Reasoning Problems with Transformer GANs

This is the implementation of the paper Generating Symbolic Reasoning Problems with Transformer GANs.

Constructing training data for symbolic reasoning domains is challenging: On the one hand existing instances are typically hand-crafted and too few to be trained on directly, on the other hand synthetically generated instances are often hard to evaluate in terms of their meaningfulness.

We provide a GAN and a Wasserstein GAN equipped with Transformer encoders to generate sensible and challenging training data for symbolic reasoning domains. Even without autoregression, the GAN models produce syntactically correct problem instances. The generated data can be used as a substitute for real training data, and, especially, the training data can be generated from a real data set that is too small to be trained on directly.

For example, the models produced the following correct mathematical expressions:

and the following correct Linear-time Temporal Logic (LTL) formulas used in verification:

Installation

The code is shipped as a Python package that can be installed by executing

pip install -e .

in the impl directory (where setup.py is located). Python version 3.6 or higher is required. Additional dependencies such as tensorflow will be installed automatically. To generate datasets or solve instances immediately after generation, the LTL satisfiability checking tool aalta is required as binary. It can be obtained from bitbucket (earliest commit in that repository). After compiling, ensure that the binary aalta resides under the bin folder.

Datasets

A zip file containing our original datasets can be downloaded from here. Unpack its contents to the datasets directory.

Dataset generation

Alternatively, datasets can be generated from scratch. The following procedure describes how to construct a dataset similar to the main base dataset (LTLbase):

First, generate a raw dataset by

python -m tgan_sr.data_generation.generator -od datasets/LTLbase --splits all_raw:1 --timeout 2 -nv 10 -ne 1600000 -ts 50 --log-each-x-percent 1 --frac-unsat None

(possibly rename to not override the supplied dataset). Enter the newly created directory.

Optional: Visualize the dataset (like Figures 5 and 6 in the paper)

python -m tgan_sr.utils.analyze_dataset all_raw.txt formula,sat

To filter the dataset for duplicates and balance classes per size

python -m tgan_sr.utils.update_dataset all_raw.txt unique - | python -m tgan_sr.utils.update_dataset - balance_per_size all_balanced.txt

Optional: Calculate relaxed satisfiability

python -m tgan_sr.utils.update_dataset all_balanced.txt relaxed_sat all_balanced_rs.txt

Optional: Visualize the dataset (like Figures 7 and 8 in the paper)

python -m tgan_sr.utils.analyze_dataset all_balanced_rs.txt formula,sat+relaxed

Split the data into training and validation sets

python -m tgan_sr.utils.update_dataset all_balanced_rs.txt shuffle+split=train:8,val:1,test:1

Experiments (training)

The folder configs contains JSON files for each type of experiment in the paper. Settings for different hyperparameters can be easily adjusted.

A model can be trained like this:

python -m tgan_sr.train.gan --run-name NAME --params-file configs/CONFIG.json

During training, relevant metrics will be logged to train_custom in the run's directory and can be viewed with tensorboard afterwards.

A list of all configurations and corresponding JSON files:

  • Standard WGAN: wgan_gp10_nl6-4_nc2_bs1024.json
  • Standard GAN: gan_nl6-4_nc2_bs1024.json
  • different σ for added noise: add parameter "gan_sigma_real" and assign desired value
  • WGAN on 10K-sized base dataset: n10k_wgan_gp10_nl6-4_nc2_bs512.json
  • Sample data from the trained WGAN: sample_n10k_wgan_gp10_nl6-4_nc2_bs512.json (ensure the "load_from" field matches your trained run name)
  • Classifier on default dataset: class_nl4_bs1024.json
  • Classifier on generated dataset: class_Generated_nl4_bs1024.json
  • WGAN with included classifier: wgan+class_nl6-3s1_nc2_bs1024.json
  • WGAN with absolute uncertainty objective: wgan+class+uncert-abs_nl6-3s1_nc2_bs1024.json (ensure the "looad_from" field matches your pre-trained name)
  • WGAN with entropy uncertainty objective: wgan+class+uncert-entr_nl6-3s1_nc2_bs1024.json (ensure the "looad_from" field matches your pre-trained name)
  • Sample data from the trained WGAN with entropy uncertainty objective: sample_wgan+class+uncert-entr_nl6-3s1_nc2_bs1024.json (ensure the "load_from" field matches your trained run name)

Evaluation

To test a trained classifier on an arbitrary dataset (validation):

python -m tgan_sr.train.gan --run-name NAME --test --ds-name DATASET_NAME

The model will be automatically loaded from the latest checkpoint in the run's directory.

How to Cite

@article{TGAN-SR,
    title = {Generating Symbolic Reasoning Problems with Transformer GANs},
    author = {Kreber, Jens U and Hahn, Christopher},
    journal = {arXiv preprint},
    year = {2021}
}
Owner
Reactive Systems Group
Saarland University
Reactive Systems Group
A general-purpose, flexible, and easy-to-use simulator alongside an OpenAI Gym trading environment for MetaTrader 5 trading platform (Approved by OpenAI Gym)

gym-mtsim: OpenAI Gym - MetaTrader 5 Simulator MtSim is a simulator for the MetaTrader 5 trading platform alongside an OpenAI Gym environment for rein

Mohammad Amin Haghpanah 184 Dec 31, 2022
RLDS stands for Reinforcement Learning Datasets

RLDS RLDS stands for Reinforcement Learning Datasets and it is an ecosystem of tools to store, retrieve and manipulate episodic data in the context of

Google Research 135 Jan 01, 2023
External Attention Network

Beyond Self-attention: External Attention using Two Linear Layers for Visual Tasks paper : https://arxiv.org/abs/2105.02358 Jittor code will come soon

MenghaoGuo 357 Dec 11, 2022
Implementation of CVPR'2022:Reconstructing Surfaces for Sparse Point Clouds with On-Surface Priors

Reconstructing Surfaces for Sparse Point Clouds with On-Surface Priors (CVPR 2022) Personal Web Pages | Paper | Project Page This repository contains

151 Dec 26, 2022
Retinal Vessel Segmentation with Pixel-wise Adaptive Filters (ISBI 2022)

Official code of Retinal Vessel Segmentation with Pixel-wise Adaptive Filters and Consistency Training (ISBI 2022)

anonymous 14 Oct 27, 2022
PyTorch implementation of Decoupling Value and Policy for Generalization in Reinforcement Learning

PyTorch implementation of Decoupling Value and Policy for Generalization in Reinforcement Learning

48 Dec 08, 2022
Music Source Separation; Train & Eval & Inference piplines and pretrained models we used for 2021 ISMIR MDX Challenge.

Introduction 1. Usage (For MSS) 1.1 Prepare running environment 1.2 Use pretrained model 1.3 Train new MSS models from scratch 1.3.1 How to train 1.3.

Leo 100 Dec 25, 2022
Feature board for ERPNext

ERPNext Feature Board Feature board for ERPNext Development Prerequisites k3d kubectl helm bench Install K3d Cluster # export K3D_FIX_CGROUPV2=1 # use

Revant Nandgaonkar 16 Nov 09, 2022
PyTorch implementation of EGVSR: Efficcient & Generic Video Super-Resolution (VSR)

This is a PyTorch implementation of EGVSR: Efficcient & Generic Video Super-Resolution (VSR), using subpixel convolution to optimize the inference speed of TecoGAN VSR model. Please refer to the offi

789 Jan 04, 2023
The Fundamental Clustering Problems Suite (FCPS) summaries 54 state-of-the-art clustering algorithms, common cluster challenges and estimations of the number of clusters as well as the testing for cluster tendency.

FCPS Fundamental Clustering Problems Suite The package provides over sixty state-of-the-art clustering algorithms for unsupervised machine learning pu

9 Nov 27, 2022
Code for WSDM 2022 paper, Contrastive Learning for Representation Degeneration Problem in Sequential Recommendation.

DuoRec Code for WSDM 2022 paper, Contrastive Learning for Representation Degeneration Problem in Sequential Recommendation. Usage Download datasets fr

Qrh 46 Dec 19, 2022
A collection of inference modules for fastai2

fastinference A collection of inference modules for fastai including inference speedup and interpretability Install pip install fastinference There ar

Zachary Mueller 83 Oct 10, 2022
YoloAll is a collection of yolo all versions. you you use YoloAll to test yolov3/yolov5/yolox/yolo_fastest

官方讨论群 QQ群:552703875 微信群:15158106211(先加作者微信,再邀请入群) YoloAll项目简介 YoloAll是一个将当前主流Yolo版本集成到同一个UI界面下的推理预测工具。可以迅速切换不同的yolo版本,并且可以针对图片,视频,摄像头码流进行实时推理,可以很方便,直观

DL-Practise 244 Jan 01, 2023
AI pipelines for Nvidia Jetson Platform

Jetson Multicamera Pipelines Easy-to-use realtime CV/AI pipelines for Nvidia Jetson Platform. This project: Builds a typical multi-camera pipeline, i.

NVIDIA AI IOT 96 Dec 23, 2022
Cross-Document Coreference Resolution

Cross-Document Coreference Resolution This repository contains code and models for end-to-end cross-document coreference resolution, as decribed in ou

Arie Cattan 29 Nov 28, 2022
SOTR: Segmenting Objects with Transformers [ICCV 2021]

SOTR: Segmenting Objects with Transformers [ICCV 2021] By Ruohao Guo, Dantong Niu, Liao Qu, Zhenbo Li Introduction This is the official implementation

186 Dec 20, 2022
Fast (simple) spectral synthesis and emission-line fitting of DESI spectra.

FastSpecFit Introduction This repository contains code and documentation to perform fast, simple spectral synthesis and emission-line fitting of DESI

5 Aug 02, 2022
Pytorch implementation of Learning with Opponent-Learning Awareness

Pytorch implementation of Learning with Opponent-Learning Awareness using DiCE

Alexis David Jacq 82 Sep 15, 2022
Unofficial implementation of "TTNet: Real-time temporal and spatial video analysis of table tennis" (CVPR 2020)

TTNet-Pytorch The implementation for the paper "TTNet: Real-time temporal and spatial video analysis of table tennis" An introduction of the project c

Nguyen Mau Dung 438 Dec 29, 2022
A tight inclusion function for continuous collision detection

Tight-Inclusion Continuous Collision Detection A conservative Continuous Collision Detection (CCD) method with support for minimum separation. You can

Continuous Collision Detection 89 Jan 01, 2023