def calc_connections_colors(data, labels, hemis, args): # stat, conditions, w, threshold=0, threshold_percentile=0, color_map='jet', # norm_by_percentile=True, norm_percs=(1, 99), symetric_colors=True): M = data.shape[0] W = data.shape[2] if args.windows == 0 else args.windows L = int((M * M + M) / 2 - M) con_indices = np.zeros((L, 2)) con_values = np.zeros((L, W, len(args.conditions))) con_names = [None] * L con_type = np.zeros((L)) for cond in range(len(args.conditions)): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): if W > 1: con_values[ind, w, cond] = data[i, j, w, cond] elif data.ndim > 2: con_values[ind, w, cond] = data[i, j, cond] else: con_values[ind, w, cond] = data[i, j] if len(args.conditions) > 1: stat_data = utils.calc_stat_data(con_values, args.stat) else: stat_data = np.squeeze(con_values) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): con_indices[ind, :] = [i, j] con_names[ind] = '{}-{}'.format(labels[i], labels[j]) con_type[ind] = HEMIS_WITHIN if hemis[i] == hemis[j] else HEMIS_BETWEEN con_indices = con_indices.astype(np.int) con_names = np.array(con_names) if args.threshold_percentile > 0: args.threshold = np.percentile(np.abs(stat_data), args.threshold_percentile) if args.threshold >= 0: indices = np.where(np.abs(stat_data) > args.threshold)[0] # con_colors = con_colors[indices] con_indices = con_indices[indices] con_names = con_names[indices] con_values = con_values[indices] con_type = con_type[indices] stat_data = stat_data[indices] con_values = np.squeeze(con_values) if args.data_max == 0 and args.data_min == 0: data_max, data_min = utils.get_data_max_min(stat_data, args.norm_by_percentile, args.norm_percs) if args.symetric_colors and np.sign(data_max) != np.sign(data_min): data_minmax = max(map(abs, [data_max, data_min])) data_max, data_min = data_minmax, -data_minmax else: data_max, data_min = args.data_max, args.data_min con_colors = utils.mat_to_colors(stat_data, data_min, data_max, args.color_map) print(len(con_names)) return con_colors, con_indices, con_names, con_values, con_type, data_max, data_min
def calc_connections_colors(data, labels, hemis, args): # stat, conditions, w, threshold=0, threshold_percentile=0, color_map='jet', # norm_by_percentile=True, norm_percs=(1, 99), symetric_colors=True): M = data.shape[0] W = data.shape[2] if args.windows == 0 else args.windows L = int((M * M + M) / 2 - M) con_indices = np.zeros((L, 2)) con_values = np.zeros((L, W, len(args.conditions))) con_names = [None] * L con_type = np.zeros((L)) for cond in range(len(args.conditions)): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): if W > 1 and data.ndim == 4: con_values[ind, w, cond] = data[i, j, w, cond] elif data.ndim > 2: con_values[ind, w, cond] = data[i, j, cond] else: con_values[ind, w, cond] = data[i, j] if len(args.conditions) > 1: stat_data = utils.calc_stat_data(con_values, args.stat) else: stat_data = np.squeeze(con_values) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): con_indices[ind, :] = [i, j] con_names[ind] = '{}-{}'.format(labels[i], labels[j]) con_type[ind] = HEMIS_WITHIN if hemis[i] == hemis[j] else HEMIS_BETWEEN con_indices = con_indices.astype(np.int) con_names = np.array(con_names) data_max, data_min = utils.get_data_max_min(stat_data, args.norm_by_percentile, args.norm_percs) data_minmax = max(map(abs, [data_max, data_min])) if args.threshold_percentile > 0: args.threshold = np.percentile(np.abs(stat_data), args.threshold_percentile) if args.threshold > data_minmax: raise Exception('threshold > abs(max(data)) ({})'.format(data_minmax)) if args.threshold >= 0: indices = np.where(np.abs(stat_data) > args.threshold)[0] # con_colors = con_colors[indices] con_indices = con_indices[indices] con_names = con_names[indices] con_values = con_values[indices] con_type = con_type[indices] stat_data = stat_data[indices] con_values = np.squeeze(con_values) if args.data_max == 0 and args.data_min == 0: if args.symetric_colors and np.sign(data_max) != np.sign(data_min): data_max, data_min = data_minmax, -data_minmax else: data_max, data_min = args.data_max, args.data_min print('data_max: {}, data_min: {}'.format(data_max, data_min)) con_colors = utils.mat_to_colors(stat_data, data_min, data_max, args.color_map) print(len(con_names)) return con_colors, con_indices, con_names, con_values, con_type, data_max, data_min
def calc_connections_colors(data, labels, hemis, stat, w, threshold=0, threshold_percentile=0, color_map='jet', norm_by_percentile=True, norm_percs=(1, 99)): M = data.shape[0] W = data.shape[2] if w == 0 else w L = int((M * M + M) / 2 - M) con_indices = np.zeros((L, 2)) con_values = np.zeros((L, W, 2)) con_names = [None] * L con_type = np.zeros((L)) axis = data.ndim - 1 coh_stat = utils.calc_stat_data(data, stat, axis=axis) x = coh_stat.ravel() data_max, data_min = utils.get_data_max_min(x, norm_by_percentile, norm_percs) data_minmax = max(map(abs, [data_max, data_min])) for cond in range(2): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): if W > 1: con_values[ind, w, cond] = data[i, j, w, cond] else: con_values[ind, w, cond] = data[i, j, cond] stat_data = utils.calc_stat_data(con_values, stat) con_colors = utils.mat_to_colors(stat_data, -data_minmax, data_minmax, color_map) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): con_indices[ind, :] = [i, j] con_names[ind] = '{}-{}'.format(labels[i], labels[j]) con_type[ind] = HEMIS_WITHIN if hemis[i] == hemis[j] else HEMIS_BETWEEN con_indices = con_indices.astype(np.int) con_names = np.array(con_names) if threshold_percentile > 0: threshold = np.percentile(np.abs(stat_data), threshold_percentile) if threshold > 0: indices = np.where(np.abs(stat_data) >= threshold)[0] con_colors = con_colors[indices] con_indices = con_indices[indices] con_names = con_names[indices] con_values = con_values[indices] con_type = con_type[indices] print(len(con_names)) return con_colors, con_indices, con_names, con_values, con_type
def calc_connections_colors(subject, data, labels, hemis, stat, threshold=0, color_map='jet', norm_by_percentile=True, norm_percs=(1, 99)): # cm_big='YlOrRd', cm_small='PuBu', flip_cm_big=True, flip_cm_small=False): M = data.shape[0] W = data.shape[2] L = int((M * M + M) / 2 - M) # con_colors = np.zeros((L, W, 3)) con_indices = np.zeros((L, 2)) con_values = np.zeros((L, W, 2)) con_names = [None] * L con_type = np.zeros((L)) coh_stat = utils.calc_stat_data(data, stat, axis=3) x = coh_stat.ravel() data_max, data_min = utils.get_data_max_min(x, norm_by_percentile, norm_percs) data_minmax = max(map(abs, [data_max, data_min])) # sm = utils.get_scalar_map(threshold, data_max, color_map=color_map) for cond in range(2): for w in range(W): # win_colors = utils.mat_to_colors(coh[:, :, w, cond], threshold, max_x, color_map, sm) # coh_arr = utils.lower_rec_to_arr(coh[:, :, w, cond]) # win_colors = utils.arr_to_colors(coh_arr, threshold, max_x, color_map, sm) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): # con_colors[ind, w, cond, :] = win_colors[ind][:3] con_values[ind, w, cond] = data[i, j, w, cond] stat_data = utils.calc_stat_data(con_values, stat) con_colors = utils.mat_to_colors(stat_data, -data_minmax, data_minmax, color_map) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): con_indices[ind, :] = [i, j] con_names[ind] = '{}-{}'.format(labels[i].astype(str), labels[j].astype(str)) con_type[ind] = HEMIS_WITHIN if hemis[i] == hemis[j] else HEMIS_BETWEEN print(L, ind) con_indices = con_indices.astype(np.int) return con_colors, con_indices, con_names, con_values, con_type
def flatten_data(data, w=0): M = data.shape[0] L = int((M*M+M)/2-M) W = data.shape[2] if w == 0 else w values = np.zeros((L, W, 2)) for cond in range(2): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): values[ind, 0, cond] = data[i, j, cond] return values
def flatten_data(data, w=0): M = data.shape[0] L = int((M*M+M)/2-M) W = data.shape[2] if w == 0 else w values = np.zeros((L, W, 2)) for cond in range(2): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): values[ind, 0, cond] = data[i, j, cond] return values
def calc_connections_colors(data, labels, hemis, stat, w, threshold=0, threshold_percentile=0, color_map='jet', norm_by_percentile=True, norm_percs=(1, 99)): M = data.shape[0] W = data.shape[2] if w == 0 else w L = int((M * M + M) / 2 - M) con_indices = np.zeros((L, 2)) con_values = np.zeros((L, W, 2)) con_names = [None] * L con_type = np.zeros((L)) axis = data.ndim - 1 coh_stat = utils.calc_stat_data(data, stat, axis=axis) x = coh_stat.ravel() data_max, data_min = utils.get_data_max_min(x, norm_by_percentile, norm_percs) data_minmax = max(map(abs, [data_max, data_min])) for cond in range(2): for w in range(W): for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): if W > 1: con_values[ind, w, cond] = data[i, j, w, cond] else: con_values[ind, w, cond] = data[i, j, cond] stat_data = utils.calc_stat_data(con_values, stat) con_colors = utils.mat_to_colors(stat_data, -data_minmax, data_minmax, color_map) for ind, (i, j) in enumerate(utils.lower_rec_indices(M)): con_indices[ind, :] = [i, j] con_names[ind] = '{}-{}'.format(labels[i], labels[j]) con_type[ind] = HEMIS_WITHIN if hemis[i] == hemis[j] else HEMIS_BETWEEN con_indices = con_indices.astype(np.int) con_names = np.array(con_names) if threshold_percentile > 0: threshold = np.percentile(np.abs(stat_data), threshold_percentile) if threshold > 0: indices = np.where(np.abs(stat_data) >= threshold)[0] con_colors = con_colors[indices] con_indices = con_indices[indices] con_names = con_names[indices] con_values = con_values[indices] con_type = con_type[indices] print(len(con_names)) return con_colors, con_indices, con_names, con_values, con_type
def compare_coh_windows(subject, task, conditions, electrodes, freqs=((8, 12), (12, 25), (25,55), (55,110)), do_plot=False): electrodes_coh = np.load(op.join(ELECTRODES_DIR, subject, task, 'electrodes_coh_windows.npy')) meg_electrodes_coh = np.load(op.join(ELECTRODES_DIR, subject, task, 'meg_electrodes_ts_coh_windows.npy')) figs_fol = op.join(MMVT_DIR, subject, 'figs', 'coh_windows') utils.make_dir(figs_fol) results = [] for cond_id, cond in enumerate(conditions): now = time.time() for freq_id, freq in enumerate(freqs): freq = '{}-{}'.format(*freq) indices = list(utils.lower_rec_indices(electrodes_coh.shape[0])) for ind, (i, j) in enumerate(indices): utils.time_to_go(now, ind, len(indices)) meg = meg_electrodes_coh[i, j, :, freq_id, cond_id][:22] elc = electrodes_coh[i, j, :, freq_id, cond_id][:22] elc_diff = np.max(elc) - np.min(elc) meg *= elc_diff / (np.max(meg) - np.min(meg)) meg += np.mean(elc) - np.mean(meg) if sum(meg) > len(meg) * 0.99: continue data_diff = meg - elc # data_diff = data_diff / max(data_diff) rms = np.sqrt(np.mean(np.power(data_diff, 2))) corr = np.corrcoef(meg, elc)[0, 1] results.append(dict(elc1=electrodes[i], elc2=electrodes[j], cond=cond, freq=freq, rms=rms, corr=corr)) if electrodes[i]=='RAF6' and electrodes[j] == 'LOF4': #corr > 10 and rms < 3: plt.figure() plt.plot(meg, label='prediction') plt.plot(elc, label='electrode') plt.legend() # plt.title('{}-{} {} {}'.format(electrodes[i], electrodes[j], freq, cond)) # (rms:{:.2f}) plt.savefig(op.join(figs_fol, '{:.2f}-{}-{}-{}-{}.jpg'.format(rms, electrodes[i], electrodes[j], freq, cond))) plt.close() results_fname = op.join(figs_fol, 'results{}.csv'.format('_bipolar' if bipolar else '')) rmss, corrs = [], [] with open(results_fname, 'w') as output_file: for res in results: output_file.write('{},{},{},{},{},{}\n'.format( res['elc1'], res['elc2'], res['cond'], res['freq'], res['rms'], res['corr'])) rmss.append(res['rms']) corrs.append(res['corr']) rmss = np.array(rmss) corrs = np.array(corrs) pass
def compare_coh_windows(subject, task, conditions, electrodes, freqs=((8, 12), (12, 25), (25,55), (55,110)), do_plot=False): electrodes_coh = np.load(op.join(ELECTRODES_DIR, subject, task, 'electrodes_coh_windows.npy')) meg_electrodes_coh = np.load(op.join(ELECTRODES_DIR, subject, task, 'meg_electrodes_ts_coh_windows.npy')) figs_fol = op.join(MMVT_DIR, subject, 'figs', 'coh_windows') utils.make_dir(figs_fol) results = [] for cond_id, cond in enumerate(conditions): now = time.time() for freq_id, freq in enumerate(freqs): freq = '{}-{}'.format(*freq) indices = list(utils.lower_rec_indices(electrodes_coh.shape[0])) for ind, (i, j) in enumerate(indices): utils.time_to_go(now, ind, len(indices)) meg = meg_electrodes_coh[i, j, :, freq_id, cond_id][:22] elc = electrodes_coh[i, j, :, freq_id, cond_id][:22] elc_diff = np.max(elc) - np.min(elc) meg *= elc_diff / (np.max(meg) - np.min(meg)) meg += np.mean(elc) - np.mean(meg) if sum(meg) > len(meg) * 0.99: continue data_diff = meg - elc # data_diff = data_diff / max(data_diff) rms = np.sqrt(np.mean(np.power(data_diff, 2))) corr = np.corrcoef(meg, elc)[0, 1] results.append(dict(elc1=electrodes[i], elc2=electrodes[j], cond=cond, freq=freq, rms=rms, corr=corr)) if False: #do_plot and electrodes[i]=='RPT7' and electrodes[j] == 'RPT5': #corr > 10 and rms < 3: plt.figure() plt.plot(meg, label='pred') plt.plot(elc, label='elec') plt.legend() # plt.title('{}-{} {} {}'.format(electrodes[i], electrodes[j], freq, cond)) # (rms:{:.2f}) plt.savefig(op.join(figs_fol, '{:.2f}-{}-{}-{}-{}.jpg'.format(rms, electrodes[i], electrodes[j], freq, cond))) plt.close() results_fname = op.join(figs_fol, 'results{}.csv'.format('_bipolar' if bipolar else '')) rmss, corrs = [], [] with open(results_fname, 'w') as output_file: for res in results: output_file.write('{},{},{},{},{},{}\n'.format( res['elc1'], res['elc2'], res['cond'], res['freq'], res['rms'], res['corr'])) rmss.append(res['rms']) corrs.append(res['corr']) rmss = np.array(rmss) corrs = np.array(corrs) pass