zxlzr/autonlp

WAIC AutoNlp

AutoNlp

AutoNLP-WAIC2019

AutoDL competition introduction:NeurIPS 2019 AutoDL Challenges

Team: Upwind_flys Rank: Second place

Methods:

Our algorithm process data and select models automatically, model lib contains Character-based model, word-based model, which can be selected according to data meta-feature. Then algorithm automatically select early stop strategy and restore weights based on the Information of feedback simulation.

Document description:

Code Framework is AutoNlp-WAIC2019 starting kit
AutoDL_ingestion_program/: The code and libraries used on Codalab to run your submission.
AutoDL_scoring_program/: The code and libraries used on Codalab to score your submission.
AutoDL_sample_code_submission/: An example of code submission you can use as template.
AutoDL_sample_data/: Some sample data to test your code before you submit it.

Main python module:

run_local_test.py: A python script to simulate the runtime in codalab
model.py: Implementation of our algorithm and logics
data_manager.py: Data processing related module
model_manager.py: Automatic model generation from model library

Run the project locally:

python run_local_test.py -dataset_dir=./AutoDL_sample_data/DEMO -code_dir=./AutoDL_sample_code_submission

Experiment Results:

metrics	O1	O2	O3	O4	O5
ALC	0.8139	0.9277	0.8053	0.9758	0.8870
2AUC-1	0.8168	0.9723	0.8345	0.9966	0.9447

Other related work:

Our work in AutoML and meta-learning fields: Efficient Automatic Meta Optimization Search for Few-Shot Learning