tmlearning 1.1.0

A Python library to create and use machine learning TrackMania AIs.

TMLearner

tmlearner is a library that allows for training and using custom TrackMania AIs with machine learning. It works with any version of TrackMania - or any game that uses only arrow keys, this library trains a TrackMania bot to play just like your style of driving, by feeding the neural network screenshots to predict the next keys to press Note: tmlearner must be run with administrator privileges for the keyboard module to work when tabbed out, which is essential for use.

Features

Feature	Support
Screenshot input	✅Supported
Other data (yaw, speed) input	🟥Not supported
Digital output	✅Supported
Analog output	🟨Coming soon

Documentation

The tmlearner library has two main files, the main TMLearner class, and two functions named wasd_key_test and arrow_key_test for testing input functionality.

Input Testing

Execute the wasd_key_test and arrow_key_test functions depending on the keys that you prefer to test as input. Then, press some keys to test. Example usage of the test functions:

>>> from tmlearner import wasd_key_test, arrow_key_test
>>> wasd_key_test()
Press some WASD keys.
*User presses keys*
W pressed
A pressed
D pressed
W pressed
S pressed
S pressed
KeyboardInterrupt
>>> arrow_key_test()
*User pressed keys*
up pressed
left pressed
down pressed
up pressed
right pressed
down pressed
down pressed

TMLearner Class

Initialization

The TMLearner class is the main class used to create datasets, train models, and test models. Example usage of the TMLearner class:

>>> from tmlearner import TMLearner
>>> bot = TMLearner()

The various parameters are shown here:

Parameter	Description	Type
name	The name of the bot. The files will be named with this name.	str
keys	Should be "WASD" or "ARROW". Which set of keys you will be pressing as input.	str
data_capture_interval	The delay in seconds to capture frames when creating a dataset.	int, float
exec_capture_interval	The delay in seconds to capture frames when using the model.	int, float
save_frequency	The frequency to save the database when creating a database. Set to None for no auto-saving.	int, None
img_size	The size images are scaled to for training	tuple
nn_arch	The architecture of the neural network.	tuple
nn_random_state	The random state of the neural network. Use for reproducibility.	int
nn_test_percentage	The percentage of the dataset to hold for testing.	float
verbose	Whether to allow printing debug data. Some critical information will be printed regardless.	bool

Deleting the Files

TMLearner.delete_database method will delete the database file, the image folder, and the model file. Coming soon: Delete database but keep model. Example usage:

>>> bot.delete_database()
Deleted all of tmlearner_bot data.

Dataset Creation

TMLearner.create_database creates/adds to the existing database. You will be prompted to press [ENTER], and a 5 second countdown will commence. When it ends, data (screenshots + keys) will start being recorded. Hold the stop key, 'z', for about twice your data_capture_interval to stop and save. Example usage:

>>> bot.create_database()
Press [ENTER] to begin recording.
You will have 5 seconds to switch to TM.
Hold 'x' for about 0.5 seconds to stop.
DO NOT keyboard interrupt.

*user presses [ENTER]*
5... 4... 3... 2... 1...
GO!

Capture 1 complete. State: (False, False, True, False)
Capture 2 complete. State: (False, True, True, False)
Capture 3 complete. State: (False, True , True, False)
Capture 4 complete. State: (True, False, False, False)
*user presses 'z'*
Stopping and saving.

The dataset will be created and saved.

Model Training

TMLearner.train_model function will train the neural network with the specified settings and save in a local file. Example usage:

>>> bot.train_model()
*various debug message showing progress (if verbose), and final print showing accuracy*

Using the Model

TMLearner.run_model will actually use the model, giving the user 5 seconds to switch to the TrackMania window, before taking screenshots, running them through the model, and executing the output. The TMLearner.exec_capture_interval parameter will control the frequency that images are taken and processed. Example usage:

>>> bot.run_model()
Press [ENTER] to begin. You will have 5 seconds to switch to TM.

*user presses [ENTER]*
5... 4... 3... 2... 1...
GO! (Ctrl+C to stop)

Prediction: (False, False, True, False)
Prediction: (False True, True, False)
Prediction: (True, False, False, True)
*user pressed Ctrl+C*
Stopping. Releasing all keys.
Done

Version History

1.1.0

• First release.