search

kkuette / TradzQAI

Trading environnement for RL agents, backtesting and training.
Apache License 2.0
164 stars 47 forks source link

how to perform test using the trained model #2

Closed lamhk closed 5 years ago

lamhk commented 5 years ago

Hi kkuette,

After the training, how can I test a new datasets using the trained model? Thanks.

Regards, lamhk

kkuette commented 5 years ago

When you train a model, a save/ directory is created. The config, the logs and the model are saved in the model name (like DQN) directory in save/. the path should be ./save/model_name_id. if you use the same settings as the settings of one of the model in the save directory it will automatically use the saved model. In the default config directory (./config/), in the file environnement.json, you can change the line "data_directory": "data/" by "data_directory":"mydirectory". It take all files in all directory within the provided directory. Filepath used for training and eval are saved in <mode>_data_datetime.txt in your save directory. Once you change the data_directory path you can launch the eval by using py run.py -m eval.

lamhk commented 5 years ago

Hi kuette,

when I use "-g on" to perform training, I have the following error "Traceback (most recent call last): File "/home/lamhk/TradzQAI-master/GUI/main_window.py", line 170, in lock_error self.Build_Primary_Window() File "/home/lamhk/TradzQAI-master/GUI/main_window.py", line 590, in Build_Primary_Window self.overview = OverviewWindow(self.main_tab, self.env) File "/home/lamhk/TradzQAI-master/GUI/overview_window.py", line 26, in init GB.addWidget(self.Agent_Inventory_Init(env), 0, 2, 2, 1) File "/home/lamhk/TradzQAI-master/GUI/overview_window.py", line 75, in Agent_Inventory_Init GBox.setFixedSize(h,w) UnboundLocalError: local variable 'w' referenced before assignment ".

Thanks.

kkuette commented 5 years ago

It should be good right now, but keep in mind that the gui is kind of old, it slow down the whole process du to it's recurrent update. I will update it when i'll finish the cbpro API integration.

lamhk commented 5 years ago

Hi kkuette, thanks for your quick response. I tried to perform the training by using the default command line or add -m train. But the result always shows "0it [00:00, ?it/s]" in a single line. Anything wrong with my setting as I can see data DAX30... in the data directory and the configuration file also shows "data/" as the data_directoty. Thanks.

Regards, lamhk

kkuette commented 5 years ago

I've looked into it, and i've fixed a problem in worker, i've also just commited it. I've also added some settings ;)

lamhk commented 5 years ago

does it mean I need to download the whole set again? Thanks.

PS: "worker" means?

kkuette commented 5 years ago

Now you have to pull from your git directory. And the worker is a thread that loop all datasets, episodes, steps, located in TradzQAI/core/worker/. If you have any other problems, please open a new issue.