text_gcn.pytorch's Introduction

text_gcn.pytorch

This project implements "Graph Convolutional Networks for Text Classification. Yao et al. AAAI2019." in PyTorch.

This implementation highly based on official code yao8839836/text_gcn.

Require

Python 3.6
PyTorch 1.0

Running training and evaluation

cd ./preprocess
Run python remove_words.py <dataset>
Run python build_graph.py <dataset>
cd ..
Run python train.py <dataset>
Replace <dataset> with 20ng, R8, R52, ohsumed or mr

Visualization

R8

text_gcn.pytorch's Issues

word vector map not found?

In the preprocess/build_graph.py , word_vector_map is always a None dic.
There is some annotations about how to built a word_vector _map. But It seems not your final idea?
However, run build_graph.py will not generate word nodes because the word_vector_map is None.

Softmax

Why not use softmax for gcnlayer2?

TypeError: cuda() missing 1 required positional argument: 'self'

when i run python train.py mr, error reporting
Traceback (most recent call last):
File "train.py", line 82, in
model_func = model_func.cuda()
TypeError: cuda() missing 1 required positional argument: 'self'
Do you know how to solve the error?

Node Features Initialization

I didn't understand the point of generating node vectors in ./preprocess/build_graph.py when using an Identity matrix as node features in ./train.py. Please correct me if I am wrong.

run meets wrong and CUDA out of memory.

1、File "train.py", line 82, in
model_func = model_func.cuda()
2、 CUDA out of memory. Tried to allocate 28.27 GiB (GPU 0; 15.77 GiB total capacity; 47.02 MiB already allocated;

why use so much cuda memeroy?