from __future__ import division import lasagne import nolearn.lasagne import pecdeeplearn as pdl import data_path import time import numpy as np # Create an experiment object to keep track of parameters and facilitate data # loading and saving. exp = pdl.utils.Experiment(data_path.get()) exp.create_experiment('single_cube_dense_three_landmark_targeted_more') exp.add_param('num_training_volumes', 45) exp.add_param('max_points_per_volume', 25000) exp.add_param('margins', (12, 12, 12)) exp.add_param('local_patch_shape', [25, 25, 25]) exp.add_param('local_patch_input_shape', [25 * 25 * 25]) exp.add_param('landmark_1', 'Sternal angle') exp.add_param('landmark_2', 'Left nipple') exp.add_param('landmark_3', 'Right nipple') exp.add_param('join_dense1_num_units', 25000) exp.add_param('batch_size', 5000) exp.add_param('update_learning_rate', 0.001) exp.add_param('update_momentum', 0.9) exp.add_param('max_epochs', 100) exp.add_param('validation_prop', 0.2) exp.add_param('prediction_margins', (40, 40, 40)) # List and load all volumes. vol_list = exp.list_volumes()
from __future__ import division import lasagne import nolearn.lasagne import pecdeeplearn as pdl import data_path import time import numpy as np # Create an experiment object to keep track of parameters and facilitate data # loading and saving. exp = pdl.utils.Experiment(data_path.get()) exp.create_experiment('single_s_dense_six_landmark') exp.add_param('num_training_volumes', 45) exp.add_param('max_points_per_volume', 25000) exp.add_param('margins', (12, 12, 0)) exp.add_param('local_patch_shape', [1, 25, 25]) exp.add_param('local_patch_input_shape', [25 * 25]) exp.add_param('landmark_1', 'Sternal angle') exp.add_param('landmark_2', 'Left nipple') exp.add_param('landmark_3', 'Right nipple') exp.add_param('landmark_4', 'Left humerus ball') exp.add_param('landmark_5', 'Right humerus ball') exp.add_param('landmark_6', 'Spinal cord') exp.add_param('join_dense1_num_units', 972) exp.add_param('batch_size', 5000) exp.add_param('update_learning_rate', 0.001) exp.add_param('update_momentum', 0.9) exp.add_param('max_epochs', 100) exp.add_param('validation_prop', 0.2)