oleges1 / code-completion

pytorch version of code completion with neural attention and pointer networks

MIT License

12 stars 6 forks source link

code-completion natural-language-processing pointer-networks python pytorch

readme

Code-Completion

Pytorch version of code completion with neural attention and pointer networks

TO DO LIST:

[ ] refactor preprocessing code
[ ] add python to AST code
[ ] config for preprocessing
[ ] join training for type

Requirments list:

python3 >= 3.6
torch >= 1.2, or tensorboadX for earlier versions
pyyaml

Instruction:

run python3 preprocess.py for preprocessing
run CUDA_VISIBLE_DEVICES=id python3 train.py --config=path/to/config.yml for training with specified config, list of available configscan be found at configs folder

Results for python value prediction (acc@1):

model	vocab_size 1k	vocab_size 10k	vocab_size 50k
simple_lstm	66.33	65.7	61.68, 1 epoch
attn_lstm	64.95	65.77	63.15, 1 epoch
pointer_mixture	66.62	67.05	65.3, 3 epochs

Examples:

Here will be examples of code generation