# READ ME

## packages needed

* python 3.6.4
* torch 1.4.0
* numpy 1.17

## data preparation

1. download data from https://tianchi.aliyun.com/dataset/dataDetail?dataId=52424
2. put all files in data_open_source directory
3. unzip all *.pv.*.zip
4. sample 10% of each *.pv.*.txt and named as ${DATE}.sample.pv.txt, e.g. sample 10% of 2019-05-17.pv.txt and name the result as 2019-05-17.sample.pv.txt

## model training

1. change the configurations in config.py
    * set the PUBID (publisher ID) of training dataset
    * TRAIN_DATE_PERIOD for the date of training
        * e.g. 'TRAIN_DATE_PERIOD = ["2019-05-17"]' means fetching the demands in 2019-05-18 and impressions in 2019-05-17.
2. run `python3 main.py`
3. the logs are stored in logs/ and the models are stored in models/

## model testing

1. change the configurations in config.py
    * set IS_TESTING to True
    * set IS_TESTING_FLAG to the model directory name in models/, e.g., "260_0.98"
    * set the PUBID (publisher ID) of the testing dataset
    * set the testing date in TRAIN_DATE_PERIOD.
        * e.g. 'TRAIN_DATE_PERIOD = ["2019-05-17"]' means fetching the demands and impressions in 2019-05-17.