Lyusungwon / CCM-pytorch

Pytorch implementation of 'Commonsense Knowledge Aware Conversation Generation with Graph Attention'

CCM-pytorch

Pytorch implementation of 'Commonsense Knowledge Aware Conversation Generation with Graph Attention'

Preparation

1. Download data from here and unzip at 'data' folder

2. Change file extension of {train, valid, test}.txt to .jsonl

3. Divide jsonl files into smaller files under '{train, valid, test}set_pieces' folder

   e.g. split -l 10000 trainset.jsonl trainset_pieces/piece_, and set: args.init_chunk_size = 10000

4. Replace 'glove.840B.300d.txt' under the 'data' folder with the real file holding pretrained weights

5. pip install -r requirements.txt

Storing ConceptNet triples with RedisGraph

1. Install and build Redis and RedisGraph

2. Open redis-server and load RedisGraph module:

   redis-server --loadmodule /path/to/module/src/redisgraph.so

3. python graph.py will store triples on your RAM

4. After all triples are stored, redis-cli bgsave

5. [Making AOF] For safety, make a backup of your latest dump.rdb file and transfer this backup to a safe place; then redis-cli config set appendonly yes; redis-cli config set save ""