ggnn-seq2seq

Synthesizing Programs from Logical Constraints using Deep Neural Networks. Modeled the constraints in First Order Logic using Gated Graph Neural Network. Learned the Embedding in a Multi-Modal setup. Finally, used a GRU Decoder to generate the corresponding programs to satisfy the constraints.

Requirements

• PyTorch + Cuda
• torchtext
• nltk
• typing
• time
• math
• json
• numpy
• matplotlib

Data

• Data Generation Pipeline was built upon [https://github.com/rrlcs/random-constraint-generation](Random Constraint Generation)

Download Dataset

cd data
Download from https://drive.google.com/file/d/1sfh4jp2YqH-A9_Iy8zF4QHzNCVjzfmWx/view?usp=sharing

Train the model

python3 train.py

Evaluate the model

python3 evaluate.py