Skip to content

quanthsu/FNP

Repository files navigation

FNP task1

Binary classification task

Text section contains a causal relation => 1, Otherwise => 0

Unbalanced dataset, only about 7% 1s

10837 training data & 2710 testing data

Submit your result on Kaggle

Read data descriptions and download the dataset from kaggle

Code

  1. preprocess.py: Separate the dataset into the training set, validation set, and test set.

  2. Train the models via baseline.py with BERT & Focal Loss and generate the csv file pred.csv.

  3. Generate pseudo-label training data via create_pseudo.py .

  4. Train the models via xlnet_pseudo.py with XLNet, Focal Loss & pseudo-label.

  5. Submit the result final_preditcion.csv.

Experiment environment

  • OS : Debian 9.11
  • Python 3.6
  • GPU : TITAN RTX (24G RAM)
  • CPU : Core i7-9800X

About

code for FinCausal 2020 tasks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages