# 'A Source Code of Bandit and Delayed Feedback in Online Structured Prediction'
This is a research code of Bandit and Delayed Feedback in Online Structured Prediction. 

To run the experiments the following python packages are required:
- pandas
- numpy
- matplotlib
- cvxpy
- itertools

Please move to the neurips_code folder before running the file.

## Experiments of MNIST
- To run experiments with MNIST, please run Run_mnist.py.
- To generate images for the MNIST experiment, please run make_image_mnist.py.

## Experiments of Synthetic data
- To run experiments with synthetic data, please run RunSynthetic.py.
- To generate images for the synthetic data experiment, please run make_image.py.

## Experiments of multilabel
- To run experiments, please run Run_multilabel_fixed.py.
- To generate images for the experiment, please run make_image_multilabel.py.

## Note
- We used the source code of loss.py, runclass.py, RunExperiment.py, RunSynthetic.py, scalefreeMD.py, synsep.py and compared.py from  Beyond Bandit Feedback in Online Multiclass Classification (Hoeven et.al. 2021.) (https://proceedings.neurips.cc/paper/2021/hash/6e79ed05baec2754e25b4eac73a332d2-Abstract.html) and partially modifiied for use.

