Push master configuration

examples/benchmarks/MASTER/README.md

@@ -5,9 +5,17 @@ paper: [MASTER: Market-Guided Stock Transformer for Stock Price Forecasting](htt
 
 codes: [https://github.com/SJTU-Quant/MASTER](https://github.com/SJTU-Quant/MASTER)
 
+## config
+We recommend you to use conda to config the environment and run the codes:
+> Note that you should install `torch` and by your self.
+```
+bash config.sh
+```
+
 ## run
 You can directly use the bash script to run the codes (you can set the `universe` and `only_backtest` flag in `run.sh`), this `main.py` will test the model with 3 random seeds:
 ```
+conda activate MASTER
 bash run.sh
 ```
 <!-- or you can just directly use `qrun` tp run the codes (note that you should modify your `qlib`, since we add or modify some files in `qlib/contrib/data/dataset.py`, `qlib/data/dataset/__init__.py`, `qlib/data/dataset/processor.py` and `qlib/contrib/model/pytorch_master.py`):

examples/benchmarks/MASTER/config.sh

@@ -0,0 +1,19 @@
+# create a new conda environment named `MASTER`
+conda create -n MASTER python=3.8
+eval "$(conda shell.bash hook)"
+conda activate MASTER
+
+# install `qlib`
+pip install pyqlib
+pip install -r requirements.txt
+
+# the following codes copy the `*.so` files in the installed `qlib` package to our customizable `qlib`
+path=$(which conda)
+parent_path=$(dirname $path)
+parent_path=$(dirname $parent_path)
+lib_path="$parent_path/envs/MASTER/lib/python3.8/site-packages/qlib/data/_libs"
+echo $lib_path
+find $lib_path -type f -name "*.so" -exec cp {} "../../../qlib/data/_libs" \;
+
+# we directly use our customizable `qlib`
+pip uninstall pyqlib

examples/benchmarks/MASTER/main.py

@@ -104,4 +104,4 @@ def parse_args():
             pp.pprint(all_metrics)
 
     for k in all_metrics.keys():
-        print(f"{k}: {np.mean(all_metrics[k])} +- {np.std(all_metrics[k])}")
+        print(f"{k}: {np.mean(all_metrics[k])} +- {np.std(all_metrics[k])}")

examples/benchmarks/MASTER/requirements.txt

@@ -0,0 +1 @@
+chardet

examples/benchmarks/MASTER/run.sh

@@ -5,8 +5,11 @@ if [ ! -d "./backtest" ]; then
     mkdir ./backtest
 fi
 
-# set the config
+# set the market, you can choose `csi300` or `csi500`
 universe=csi300
+
+# set the flag: whether only backtest. 
+# If choose 'true', we will directly use the trained model and do backtesting; If choose 'false', the script will first train then backtest.
 only_backtest=false
 
 sed -i "s/csi.../$universe/g" workflow_config_master_Alpha158.yaml