rewards = colorPerEpisode(training_data['episode_starts'])
# Compute Ground Truth Correlation
gt_corr, gt_corr_mean = plotCorrelation(states_rewards, ground_truth, target_positions,
only_print=args.print_corr)
result_dict = {
'gt_corr': gt_corr.tolist(),
'gt_corr_mean': gt_corr_mean
}
# Write the results in a json file
log_folder = os.path.dirname(args.input_file)
with open("{}/gt_correlation.json".format(log_folder), 'w') as f:
json.dump(result_dict, f)
else:
plotRepresentation(states_rewards['states'], rewards, cmap=cmap)
if not args.print_corr:
getInputBuiltin()('\nPress any key to exit.')
elif args.data_folder != "":
print("Plotting ground truth...")
training_data, ground_truth, true_states, _ = loadData(args.data_folder)
rewards = training_data['rewards']
name = "Ground Truth States - {}".format(args.data_folder)
if args.color_episode:
rewards = colorPerEpisode(training_data['episode_starts'])
if args.plot_against:
plotAgainst(true_states, rewards, cmap=cmap)
elif args.pretty_plot_against:
parser.add_argument('--no-display-plots',
action='store_true',
default=False,
help='disables live plots of the representation learned')
parser.add_argument('--data-folder',
type=str,
default="",
help='Dataset folder',
required=True)
parser.add_argument('--training-set-size',
type=int,
default=-1,
help='Limit size of the training set (default: -1)')
parser.add_argument('--state-dim', type=int, default=3, help='State dimension')
input = getInputBuiltin()
args = parser.parse_args()
DISPLAY_PLOTS = not args.no_display_plots
plot_script.INTERACTIVE_PLOT = DISPLAY_PLOTS
args.data_folder = parseDataFolder(args.data_folder)
args.method = "pca"
log_folder = "logs/{}/baselines/{}".format(args.data_folder,
getModelName(args))
createFolder(log_folder, "{} folder already exist".format(args.method))
folder_path = '{}/NearestNeighbors/'.format(log_folder)
createFolder(folder_path, "NearestNeighbors folder already exist")
saveExpConfig(args, log_folder)
print('Log folder: {}'.format(log_folder))