def create_a_template(id):
"""
create template
:param id: template id
:return: a DataUnit instance
"""
if id == 0:
# template 0, used in ISMRM2017 abstract
# new and setting
du = toolboxes.DataUnit()
du._secured_data['if_random_neural_parameter'] = False
du._secured_data['if_random_hemodynamic_parameter'] = False
du._secured_data['if_random_x_state_initial'] = False
du._secured_data['if_random_h_state_initial'] = False
du._secured_data['if_random_stimuli'] = True
du._secured_data['if_random_node_number'] = False
du._secured_data['if_random_stimuli_number'] = False
du._secured_data['if_random_delta_t'] = False
du._secured_data['if_random_scan_time'] = False
du._secured_data['t_delta'] = 1. / 16.
du._secured_data['t_scan'] = 5 * 60
du._secured_data['n_node'] = 3
du._secured_data['n_stimuli'] = 1
du._secured_data['A'] = np.array([[-1, 0, 0],
[0.8, -1, 0.4],
[0.4, 0.8, -1]])
du._secured_data['B'] = [np.array([[0, 0, 0],
[0, 0, 0],
[0, 0, -0.4]])]
du._secured_data['C'] = np.array([0.8, 0, 0]).reshape(3, 1)
du._secured_data['learning_rate'] = 0.1
du._secured_data['n_backpro'] = 12
du.complete_data_unit(if_show_message=False, if_check_property=False)
return du
def _run():
# print('_run() runs')
# cores = dbo.load_database(RESULT_PATH_DCM_RNN)
with open(data_path, 'rb') as f:
cores = pickle.load(f)
# for each DCM, recover fMRI signal with t_delta = 1/16
t_delta_list = [1.0 / 16]
rMSEs = []
for i in range(len(cores)):
# print('current processing: ' + str(i))
du = tb.DataUnit()
du.load_parameter_core(cores[i])
y_list = du.map('t_delta', t_delta_list, 'y')
signal_length = y_list[0].shape[0]
# y_down_sampled = du.resample_arrays(y_list, (int(signal_length / 32), du.get('n_node')))
# y_up_sampled = du.resample_arrays(y_down_sampled, (signal_length, du.get('n_node')), order=3)
y_down_sampled = [
du.resample(array, (int(signal_length / 32), du.get('n_node')))
for array in y_list
]
y_up_sampled = [
du.resample(array, (signal_length, du.get('n_node')))
for array in y_down_sampled
]
rMSE = du.compare(y_up_sampled, y_list[0])
rMSEs.append(rMSE)
# save results
output_dir = os.path.split(output_file)[0]
print('output_dir: ' + output_dir)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
with open(output_file, 'wb') as f:
pickle.dump(rMSEs, f)
def _run():
# print('_run() runs')
# cores = dbo.load_database(RESULT_PATH_DCM_RNN)
with open(data_path, 'rb') as f:
cores = pickle.load(f)
# for each DCM, recover fMRI signal with different t_delta
t_delta_list = [
1.0 / 64, 1.0 / 32, 1.0 / 16, 1.0 / 8, 1.0 / 4, 1.0 / 2
]
rMSEs = []
# for i in range(len(cores)):
for i in range(len(cores)):
# print('current processing: ' + str(i))
du = tb.DataUnit()
du.load_parameter_core(cores[i])
y_list = du.map('t_delta', t_delta_list, 'y')
y_resampled = [
du.resample(array, y_list[-1].shape) for array in y_list
]
rMSE = du.compare(y_resampled, y_resampled[0])
rMSEs.append(rMSE)
# save results
output_dir = os.path.split(output_file)[0]
print('output_dir: ' + output_dir)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
with open(output_file, 'wb') as f:
pickle.dump(rMSEs, f)
interpolate_func = interp1d(t_axis_original, spm_data['y'][:, i], kind='cubic')
spm_data['y_upsampled'][:, i] = interpolate_func(t_axis_new)
spm_data['u_upsampled'] = np.zeros((len(t_axis_new), spm_data['u'].shape[1]))
for i in range(spm_data['u'].shape[1]):
interpolate_func = interp1d(t_axis_original, spm_data['u'][:, i])
spm_data['u_upsampled'][:, i] = interpolate_func(t_axis_new)
temporal_mask = spm_data['u_upsampled'][:, 0] < 0.5
temp = copy.deepcopy(spm_data['y_upsampled'])
mean_values = np.mean(temp[temporal_mask, :], axis=0)
mean_values = np.repeat(mean_values.reshape([1, 3]), spm_data['y_upsampled'].shape[0], axis=0)
spm_data['y_upsampled'] = spm_data['y_upsampled'] - mean_values
N_SEGMENTS = mth.ceil(len(spm_data['y_upsampled']) / N_RECURRENT_STEP) * N_RECURRENT_STEP
# build DataUnit du from spm_data
du = tb.DataUnit()
du._secured_data['n_node'] = spm_data['y_upsampled'].shape[1]
du._secured_data['n_stimuli'] = spm_data['u_upsampled'].shape[1]
du._secured_data['t_delta'] = TARGET_TEMPORAL_RESOLUTION
du._secured_data['t_scan'] = du._secured_data['t_delta'] * spm_data['y_upsampled'].shape[0]
du._secured_data['n_time_point'] = spm_data['y_upsampled'].shape[0]
du._secured_data['u'] = spm_data['u_upsampled']
du._secured_data['y'] = spm_data['y_upsampled']
# initialize DataUnit du_hat, support masks should be specified here
# Wxx = np.zeros((du.get('n_node'), du.get('n_node')))
Wxx = np.eye(du.get('n_node')) - np.eye(du.get('n_node')) * du.get('t_delta')
Wxxu = [np.zeros((du.get('n_node'), du.get('n_node'))) for _ in range(du.get('n_stimuli'))]
Wxu = np.zeros((du.get('n_node'), du.get('n_stimuli')))
Wxu[0, 0] = 1. * du.get('t_delta')
print("PROJECT_DIR is set as: " + PROJECT_DIR)
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import os
import pickle
import toolboxes
#EXPERIMENT_DIR = os.path.join(PROJECT_DIR, 'experiments', 't_delta')
#INPUT_DIR = os.path.join(PROJECT_DIR, 'data', 'DB0.pkl')
# create new DCM
# new and setting
du = toolboxes.DataUnit()
du._secured_data['if_random_node_number'] = True
du._secured_data['if_random_stimuli'] = True
du._secured_data['if_random_x_state_initial'] = False
du._secured_data['if_random_h_state_initial'] = False
du._secured_data['t_delta'] = 0.25
du._secured_data['t_scan'] = 5 * 60
du.complete_data_unit(if_show_message=False)
x_axis = np.array(range(0, len(du.get('u')))) * du.get('t_delta')
plt.figure(figsize=(4, 2), dpi=300)
plt.subplot(2, 1, 1)
for i in range(du.get('n_stimuli')):
plt.plot(x_axis, du.get('u')[:, i], linewidth=0.75, alpha=0.75)
plt.ylabel('inputs')
import importlib
from scipy.fftpack import idct, dct
DATA_DIR = os.path.join(PROJECT_DIR, 'dcm_rnn', 'resources',
'SPM_data_attention.pkl')
MAX_EPOCHS = 32 * 3
CHECK_STEPS = 1
N_RECURRENT_STEP = 192
DATA_SHIFT = 2
MAX_BACK_TRACK = 16
MAX_CHANGE = 0.001
BATCH_RANDOM_DROP_RATE = 1.
TARGET_T_DELTA = 1. / 16
N_CONFOUNDS = 19
du = tb.DataUnit()
# load SPM attention spm_data from resources
spm_data = pickle.load(open(DATA_DIR, 'rb'))
# process spm_data,
# up sample u and y to 16 frame/second
shape = list(spm_data['u'].shape)
shape[0] = int(spm_data['TR'] / TARGET_T_DELTA * shape[0])
spm_data['u_upsampled'] = du.resample(spm_data['u'], shape, order=3)
spm_data['y_upsampled'] = du.resample(spm_data['y'], shape, order=3)
# assume the observation model is
# y = DCM_RNN(u) + Confounds * weights + noise
# Confounds are the first cosine transfer basis (19)
n_time_point = spm_data['u_upsampled'].shape[0]
confounds = idct(np.eye(n_time_point)[:, :N_CONFOUNDS], axis=0, norm='ortho')
import copy
import pandas as pd
import math as mth
from scipy.interpolate import interp1d
import scipy as sp
import scipy.io
EXPERIMENT_PATH = os.path.join(PROJECT_DIR, 'experiments',
'compare_estimation_with_simulated_data')
DATA_PATH = os.path.join(EXPERIMENT_PATH, 'data')
CORE_PATH = os.path.join(DATA_PATH, 'core.pkl')
SAVE_PATH_PKL = os.path.join(DATA_PATH, 'du_DCM_RNN.pkl')
SAVE_PATH_MAT = os.path.join(DATA_PATH, 'DCM_initial.mat')
SNR = 3
du = tb.DataUnit()
du._secured_data['if_random_neural_parameter'] = False
du._secured_data['if_random_hemodynamic_parameter'] = False
du._secured_data['if_random_x_state_initial'] = False
du._secured_data['if_random_h_state_initial'] = False
du._secured_data['if_random_stimuli'] = True
du._secured_data['if_random_node_number'] = False
du._secured_data['if_random_stimuli_number'] = False
du._secured_data['if_random_delta_t'] = False
du._secured_data['if_random_scan_time'] = False
du._secured_data['t_delta'] = 1. / 64.
du._secured_data['t_scan'] = 5 * 60
du._secured_data['n_node'] = 3
du._secured_data['n_stimuli'] = 3
du._secured_data['A'] = np.array([[-0.8, 0., 0.], [0., -0.8, 0.],