training_config.yaml

# model parameters
model_name : AutoregressiveLMU    # name of the model
load_state : ""                   # path to model state for loading pretrained models
dt : 0.005                         # nengo time step
t_delay : 0.02                    # how far to predict the future (initial guess)
neurons_per_dim : 50              # number of neurons representing each dimension
radius : 1.5                      # radius of the nengo ensemble
lmu_theta : 0.02                   # duration of the LMU delay
lmu_q : 1                         # number of factorizations per dim in LMU
predict_delta : True              # decide if the model should use the current state as a prior for next state prediction
seed : 4                          # to get reproducible neuron properties across runs

# training data parameters
data_dir : data/Mixed/Train/      # folder containing individual training files
bound : 0.19                      # if the cart ever leaves these bounds, the data is ignored
sample_freq : 50                  # cartpole data is recorded at ~50Hz
shuffle : True                    # shuffle the training data
skiprows : 0                      # initial rows to skip in the data .csv files
action_vars :                     # names of the action variables in training data
  - Q
state_vars :                      # names of the state variables in training data
  - angle_sin
  - angle_cos
  - angleD
  - position
  - positionD

# result parameters
results_dir : results/            # folder to save the plots and weights to
experiment_name : training        # sub-folder of results folder

# training process parameters
epochs : 1                       # number or epochs for training
learning_rate : 0.00005           # learning rate used in online learning
save_state_every : 200            # how often to save the weights of the model during epoch
plot_prediction_every : 200       # how often to plot a prediction curve during learning
max_samples : -1                  # just reduce training set for quick debugging
t_switch : 5.0                    # after how many seconds to switch to autoregressive mode
datapoints_per_file : 1000        # divide by sampling rate to get the duration in s