https://github.com/ngxbac/Kaggle-Jigsaw

https://github.com/ngxbac/Kaggle-Jigsaw

Last synced: 9 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/ngxbac/Kaggle-Jigsaw
• Owner: ngxbac
• Created: 2019-05-31T06:25:03.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-06-27T06:35:08.000Z (over 5 years ago)
• Last Synced: 2024-08-01T13:32:19.287Z (3 months ago)
• Language: Python
• Size: 115 KB
• Stars: 24
• Watchers: 4
• Forks: 11
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Kaggle-Jigsaw 

Our solution is described in [here](https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/97425#latest-562308)

# Preprocessing 

### Extract data for BERT/GPT2/XLNET

```bash

bash bin/extract_data.sh

```

### Extract features (11 features) 

```bash

bash bin/extract_features.sh

```

### Create targets 

```bash

bash bin/extract_target.sh

```

# Train models

```bash 

seed=17493

depth=12 #11, 12 for Bert base, 23, 24 for Bert large

maxlen=220

batch_size=32

accumulation_steps=4

model_name=bert #gpt2, xlnet

CUDA_VISIBLE_DEVICES=3 python main_catalyst.py train    --seed=$seed \

                                                        --depth=$depth \

                                                        --maxlen=$maxlen \

                                                        --batch_size=$batch_size \

                                                        --accumulation_steps=$accumulation_steps \

                                                        --model_name=$model_name

```

# Predictions 

Change the settings as same as training phase. 

Ex: 

```

seed=17493

depth=12

maxlen=220

batch_size=32

accumulation_steps=4

model_name=bert

```

Then

```bash 

python make_submission.py

```