This repo is forked from MTLSA, I modified some code in the repo to make it a baseline model for DRSA, which is the implementation of the model proposed in a AAAI'19 submitted paper.
Many thanks to the authors of MTLSA.
We have upload a tiny data sample for training and evaluation.
The full dataset for this project can be download from this link.
After download please replace the sample data in data/deep-bid-lands-data/ folder with the full data files.
Please install h5py first.
The code contains two parts, so you need to run it by following the steps here:
- step1: If you just want to run the demo, you should execute the run.m in MATLAB, and that is enough for step1. If you want to run it with full volume data, you should follow the instructions in previous section and modifiy the run.m script to use different campaign's data. It is very simple.
- step2:
As step1, if you just want to run the demo, just execute:
and it will give the AUC, log-loss and ANLP value trained on the sample data. If you want to get full volume data result, do step1 and modify
python3 util_MTLSA.pyutil_MTLSA.pyincampaign_listvariable.
Be patient, to run the code with demo data we need about 30 minutes. Much slower when running it with full volume dataset.