# `IMAGE_SIZE` is an (int, int) tuple of the image size of the first image.
# We assume that each image is of the same size as the image labelled "0"
config.IMAGE_SIZE = np.shape(imread(config.IMAGE_TEMPLATE.format(0)))
# `IMAGES` holds the original digit images.
config.IMAGES = [
cv2.cvtColor(
cv2.resize(imread(config.IMAGE_TEMPLATE.format(digit)),
dsize=(TRAINING_CONFIG.grid.render_shape)),
cv2.COLOR_RGBA2RGB) for digit in range(10)
]
# Initialise the checkpoint
if config.CONDITION == 0:
config.PROCESSOR = Stimulus
else:
LEARNING_RATE_ENCODER = TRAINING_CONFIG.learning_rate_encoder
LEARNING_RATE_DECODER = TRAINING_CONFIG.learning_rate_decoder
ENCODER_OPTIMISER = tf.keras.optimizers.Adam(
learning_rate=LEARNING_RATE_ENCODER)
DECODER_OPTIMISER = tf.keras.optimizers.Adam(
learning_rate=LEARNING_RATE_DECODER)
CHECKPOINT = tf.train.Checkpoint(
encoder=TRAINING_CONFIG.encoder,
decoder=TRAINING_CONFIG.decoder,
encoder_optimiser=ENCODER_OPTIMISER,
decoder_optimiser=DECODER_OPTIMISER,
)
try:
grid = grids[config.GRID_TYPE]()
except KeyError:
with open(config.GRID_TYPE, 'rb') as infile:
grid = pickle.load(infile)
config.GRID = grid
# We initiate the stimulus processor type.
processors = {
'direct': Stimulus,
'net': lambda image, grid: StimulusNet(image, grid, config.ENCODER),
}
config.PROCESSOR = processors[config.PROCESSOR_TYPE]
# Templates for data paths.
config.DATETIME_FORMAT = '%Y-%m-%d_%H-%M-%S'
config.DIGIT_SOUND_TEMPLATE = './data/digit-voice/{}-alt.wav'
base_dir = './data/psychophysics-sessions/'
if config.TESTING:
config.CONFIG_FILE_TEMPLATE = base_dir + 'tests/{}_{}_config.json'
config.SESSION_FILE_TEMPLATE = base_dir + 'tests/{}_{}_session.csv'
config.MOUSE_FILE_TEMPLATE = base_dir + 'tests/{}_{}_mouse.csv'
else:
config.CONFIG_FILE_TEMPLATE = base_dir + 'participants/{}_{}_config.json'
config.SESSION_FILE_TEMPLATE = base_dir + 'participants/{}_{}_session.csv'
config.MOUSE_FILE_TEMPLATE = base_dir + 'participants/{}_{}_mouse.csv'
