Train CNN Refactoring (#201)

* Seperate pytorch and mxnet imports

* Refactoring of supervised training

* Removed importing MXNet when training with pytorch
* Added data_loader helper method
* Removed _template suffix from config files
* Updated requirement.txt
* Updated READMEs
* Updated Dockerfile (now downloads cutechess, example NNs and calls git pull)

.gitignore

@@ -123,10 +123,6 @@ venv.bak/
 # avoid pushing any lock files
 *.~lock
 
-# avoid pushing config files
-main_config.py
-train_config.py
-
 # avoid pushing log-files generated by uci-communication
 CrazyAra-log.txt
 score-log.txt

DeepCrazyhouse/configs/README.md

@@ -1,9 +1,9 @@
 ### Configuration files for CrazyAra
 
+This is the configuration directory of CrazyAra for both supervised training and reinforcement learning.
+
 If you want to test sample MCTS predictions in [MCTS_eval_demo.ipynb](https://github.com/QueensGambit/CrazyAra/blob/master/DeepCrazyhouse/src/samples/MCTS_eval_demo.ipynb),
  then follow these steps:
-
-*   First rename `main_config_template.py` into `main_config.py`
 
 *   Specify the fields `main_config["model_architecture_dir"]` and `main_config["model_weights_dir"]` in the file
     `main_config.py` to the appropriate paths of your system. Make sure that the path has a "/" at the end of the path.

DeepCrazyhouse/src/training/README.md

@@ -2,21 +2,26 @@
 
 ### Prerequisites
 
-Make sure to have a recent [MXNet](https://mxnet.incubator.apache.org/index.html) version with CUDA support installed:
- ```bash
- pip install mxnet-cu<cuda_version>==<version_id>
-```
+Make sure to have a recent [Pytorch](https://pytorch.org/get-started/locally/) version with CUDA support installed.
 
 For supervised training you need the following [additional libraries](https://github.com/QueensGambit/CrazyAra/blob/master/DeepCrazyhouse/src/training/requirements.txt):
 
 ```bash
 pip install -r requirements.txt
 ```
 
-* zarr (chunked, compressed, N-dimensional array library)
-* numcodecs (compression codec library)
-* tqdm: (progress bar library)
-* MXBoard (logging MXNet data for visualization in TensorBoard)
+#### Training with MXNet or Gluon (deprecated)
+Make sure to have a recent [MXNet](https://mxnet.incubator.apache.org/index.html) version with CUDA support installed:
+ ```bash
+ pip install mxnet-cu<cuda_version>==<version_id>
+```
+
+You need to install the following libraries when training with MXNet:
+```bash
+    pip install -y mxboard
+    pip uninstall -y onnx
+    pip install onnx==1.3.0
+```
 
 ### Training data specification
 Specify the directories `"planes_train_dir"`, `"planes_val_dir"`, `"planes_test_dir"`, `"planes_mate_in_one_dir"` at
@@ -32,7 +37,7 @@ Use `train_cnn.ipynb` to conduct a training run.
 * <https://jupyter.org/install.html>
 
 Jupyter notebooks are displayed in a web-browser and can be launched with `jupyter notebook` from the command line. 
-After a successfull training run you can export the outputs as a html-file:  `File->Download as->Html(.html)`.
+After a successful training run you can export the outputs as a html-file:  `File->Download as->Html(.html)`.
 
 ### Tensorboard
 The [tensorboard](https://github.com/tensorflow/tensorboard) log files will be exported in `./logs` which can be viewed with tensorboard during training.

DeepCrazyhouse/src/training/requirements.txt

@@ -1,20 +1,18 @@
+torch
 numpy
 python-chess==0.23.11
 zarr
 numcodecs
 ipywidgets
 matplotlib
 tqdm
-mxboard
-tensorflow
 tensorboard
-onnx==1.3.0
+onnx
 ipytree
 setproctitle
 rtpt
 dataclasses
-mxnet
-torch
 torchsummary
 onnxruntime
 onnxsim
+fvcore