Final product of a Udemy masterclass on AI.
Note that most of this code comes from the class
The goal of this project was to create and train an ML model to drive a car along a track in the most efficient and time-saving way possible. By the end of the course, I successfully ran the model, although I wouldn't recommend training the VAE if you have less than a few tbytes of unused storage space. Also note that traning the Full World Model could take up to a month to reach peak efficiency for the car and requires a lot of storage space as well.
GENERAL STEPS:
- Build the environment (env.py)
- Generate VAE Data (extract.py)
- Train the VAE (vae_train.py)
- Generate RNN Data (series.py)
- Train the RNN (rnn_train.py)
- Train the Full World Model (train.py)
- Run the model (model.py)
This can also be found in training_process.py
The training utilizes CMA because there are less than 1000 parameters. Should you want to train on more paremeters (in the thousands), all you need to do is navigate to train.py, locate line 405 and change 'default' from 'cma' to 'pepg' as well as change 'optimizer' on line 26 from 'cma' to 'pepg.'
TO INSTALL PACKAGES FOLLOW THESE STEPS:
In your command line - conda create -n masterai python=3.6 #note that this model runs smoother on 3.6 than 3.7 conda activate masterai conda install -c install-forge tensorflow python -m pip install cma conda install -c kna pybox2d pip install pillow pip install gym == 0.9.4
ANACONDA IS RECOMMENDED When in the anaconda homescreen toggle "Applications" to material and install Spyder, then navigate to AI Masterclass files and good luck!