def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
experiment = 'fmril'
nor_method = 'no'
clf_method = 'svm'
source_id = np.arange(5)
target_id = 6
op_function = 'l1l2' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
S = len(source_id)
xs = [x[indices_sub[i]] for i in source_id]
ys = [y_target[indices_sub[i]] for i in source_id]
xt = x[indices_sub[target_id]]
yt = y_target[indices_sub[target_id]]
Cs = [sp.spatial.distance.cdist(xs[s], xs[s]) for s in range(S)]
Cs = [cs / cs.max() for cs in Cs]
ns = [len(xs[s]) for s in range(S)]
ps = [ot.unif(ns[s]) for s in range(S)]
p = ot.unif(len(xt))
lambdas = ot.unif(S)
C, T = ot.gromov.gromov_barycenters(len(xt),
Cs,
ps,
p,
lambdas,
'square_loss',
5e-4,
max_iter=100,
tol=1e-5)
ot_source = []
note = 'barycenter_{}s_{}t_{}_{}_{}_{}'.format(source_id, target_id,
experiment, nor_method,
clf_method, op_function)
print(note)
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2' 'maplin'
experiment = 'fmril'
nor_method = 'no'
clf_method = 'svm'
source_id = 3
target_id = 2
op_function = 'maplin' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
yt = y_target[indices_sub[j]]
note = 'kfold_{}s_{}t_{}_{}_{}_{}_circulaire'.format(
i, j, experiment, nor_method, clf_method, op_function)
print(note)
reg, eta = fucs.pairsubs_circular_kfold(xs,
ys,
xt,
yt,
ot_method=op_function,
clf_method=clf_method)
print('best reg and eta', reg, eta)
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
# args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2'
metric_xst = 'h'
experiment = 'fmrir'
nor_method = 'no'
clf_method = 'logis'
source_id = 1
target_id = 2
op_function = 'l1l2' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
nb_subs = len(indices_sub)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
yt = y_target[indices_sub[j]]
note = '{}s_{}t_{}_{}_{}_sinkhorn_{}_{}'.format(i, j, experiment,
nor_method, clf_method,
op_function, metric_xst)
accs = fucs.get_accuracy(xs, ys, xt, yt, clf_method='logis')
print(accs)
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
y_target = source.y_target
subjects = np.array(source.subjects)
nb_samples_each = source.nb_samples_each
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = models.split_subjects(subjects)
nb_subs = len(indices_sub)
# for i in range(len(indices_sub)-1):
# for j in range(i+1, len(indices_sub)):
i = source_id
for j in range(nb_subs):
if i != j:
scores_params = {
'source': [],
'target': [],
'ori_score': [],
'ot_score': [],
'params': []
}
note = '{}s_{}t_{}_{}_{}_skinkhorn_{}'.format(i, j, experiment,
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
# args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2'
metric_xst = "cityblock"
experiment = 'fmril'
nor_method = 'no'
clf_method = 'svm'
source_id = 1
target_id = 2
op_function = 'l1l2' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
nb_subs = len(indices_sub)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
yt = y_target[indices_sub[j]]
note = '{}s_{}t_{}_{}_{}_sinkhorn_{}_{}'.format(i, j, experiment,
nor_method, clf_method,
op_function, metric_xst)
print(note)
# cityblock sqeuclidian,
best_xst, ot_params = fucs.pairsubs_sinkhorn_lables(xs,
ys,
xt,
ot_method=op_function,
metric=metric_xst,
clf_method=clf_method)
# train on xst, test on xt
ot_accs = fucs.get_accuracy(best_xst, ys, xt, yt, clf_method=clf_method)
ot_score = np.mean(ot_accs)
print('ot source data predicts on original target, accuracy', ot_score,
ot_accs)
# train on xs, test on xt
ori_accs = fucs.get_accuracy(xs, ys, xt, yt, clf_method=clf_method)
ori_score = np.mean(ori_accs)
print('original source data predicts on original target, accuracy',
ori_score, ori_accs)
# train on xs indi-nor, test on xt indi-nor
x_base = x_indi.copy()
xs_base = x_base[indices_sub[i]]
ys_base = y_target[indices_sub[i]]
xt_base = x_base[indices_sub[j]]
yt_base = y_target[indices_sub[j]]
base_accs = fucs.get_accuracy(xs_base,
ys_base,
xt_base,
yt_base,
clf_method=clf_method)
base_score = np.mean(base_accs)
print(
'indi-normalized source predicts on indi-normalized target, accuracy',
base_score, base_accs)
pair_params = {}
pair_params['source'] = i
pair_params['target'] = j
pair_params['ori_score'] = round(ori_score, 4)
pair_params['ori_accs'] = ori_accs
pair_params['ot_accs'] = ot_accs
pair_params['ot_score'] = round(ot_score, 4)
pair_params['base_accs'] = base_accs
pair_params['base_score'] = round(base_score, 4)
pair_params['params'] = ot_params
ttest = stats.ttest_rel(base_accs, ot_accs)
pair_params['ttest'] = (ttest.pvalue, ttest.statistic)
print('pair_params', pair_params)
# if ttest.statistic > 0 and ttest.pvalue < 0.05:
# joblib.dump(pair_params, result_dir+'/big_ori/{}_pair_params.pkl'.format(note))
# elif ttest.statistic > 0 and ttest.pvalue >= 0.05:
# joblib.dump(pair_params, result_dir + '/big_ori_non_significant/{}_pair_params.pkl'.format(note))
# elif ttest.statistic <= 0 and ttest.pvalue >= 0.05:
# joblib.dump(pair_params, result_dir + '/big_ot_non_significant/{}_pair_params.pkl'.format(note))
# else:
# joblib.dump(pair_params, result_dir + '/big_ot/{}_pair_params.pkl'.format(note))
print(
'-----------------------------------------------------------------------------------------'
)
print(best_xst, xs)
# print(scores_params)
# note = '{}s_{}ts_{}_{}_{}_sinkhorn_{}_{}'.format(i, len(target_ids), experiment,
# nor_method, clf_method, op_function, metric_xst)
# joblib.dump(scores_params, result_dir + '/{}_score_params.pkl'.format(note))
# print('unsuccessful transport', unsec)
print(time.strftime('%Y-%m-%d %A %X %Z', time.localtime(time.time())))
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
args = sys.argv[1:]
experiment = args[0]
nor_method = args[1]
clf_method = args[2]
source_id = int(args[3])
target_id = int(args[4])
op_function = args[5] # 'lpl1' 'l1l2' 'maplin'
# experiment = 'fmril'
# nor_method = 'no'
# clf_method = 'svm'
# source_id = 3
# target_id = 2
# op_function = 'maplin' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
yt = y_target[indices_sub[j]]
note = '{}s_{}t_{}_{}_{}_{}_circulaire'.format(i, j, experiment,
nor_method, clf_method, op_function)
print(note)
# cityblock sqeuclidian, minkowski
reg, eta = fucs.pairsubs_circular(xs, ys, xt, yt, ot_method=op_function, clf_method=clf_method)
print('best reg and eta', reg, eta)
if op_function == 'l1l2':
transport = ot.da.SinkhornL1l2Transport(reg_e=reg, reg_cl=eta, norm='max')
elif op_function == 'lpl1':
transport = ot.da.SinkhornLpl1Transport(reg_e=reg, reg_cl=eta, norm='max')
elif op_function == 'maplin':
transport = ot.da.MappingTransport(kernel="linear", mu=reg, eta=eta, bias=True, norm='max')
else:
print('Warning: need to choose among "l1l2", "lpl1" and "maplin"', op_function)
# train on xst, test on xt
transport.fit(Xs=xs, Xt=xt, ys=ys)
xst = transport.transform(Xs=xs)
ot_accs = fucs.get_accuracy(xst, ys, xt, yt, clf_method=clf_method)
ot_score = np.mean(ot_accs)
print('train on ot source data, predict on original target, accuracy', ot_score, ot_accs)
# train on xs, test on xt
ori_accs = fucs.get_accuracy(xs, ys, xt, yt, clf_method=clf_method)
ori_score = np.mean(ori_accs)
print('train on original source data, predict on original target, accuracy', ori_score, ori_accs)
# train on xs indi-nor, test on xt indi-nor
x_base = x_indi.copy()
xs_base = x_base[indices_sub[i]]
ys_base = y_target[indices_sub[i]]
xt_base = x_base[indices_sub[j]]
yt_base = y_target[indices_sub[j]]
base_accs = fucs.get_accuracy(xs_base, ys_base, xt_base, yt_base, clf_method=clf_method)
base_score = np.mean(base_accs)
print('base: indi-normalized source predicts on indi-normalized target, accuracy', base_score, base_accs)
pair_params = {}
pair_params['source'] = i
pair_params['target'] = j
pair_params['ori_score'] = round(ori_score, 4)
pair_params['ori_accs'] = ori_accs
pair_params['ot_accs'] = ot_accs
pair_params['ot_score'] = round(ot_score, 4)
pair_params['base_accs'] = base_accs
pair_params['base_score'] = round(base_score, 4)
pair_params['params'] = (reg, eta)
ttest_base_ot = stats.ttest_rel(base_accs,ot_accs)
pair_params['ttest_base_ot'] = (ttest_base_ot.pvalue, ttest_base_ot.statistic)
ttest_ori_ot = stats.ttest_rel(ori_accs, ot_accs)
pair_params['ttest_ori_ot'] = (ttest_ori_ot.pvalue, ttest_ori_ot.statistic)
print('pair_params', pair_params)
if ttest_base_ot .statistic > 0:
joblib.dump(pair_params, result_dir+'/big_base/{}_pair_params.pkl'.format(note))
else:
joblib.dump(pair_params, result_dir + '/small_base/{}_pair_params.pkl'.format(note))
if ttest_ori_ot .statistic > 0:
joblib.dump(pair_params, result_dir+'/big_ori/{}_pair_params.pkl'.format(note))
else:
joblib.dump(pair_params, result_dir + '/small_ori/{}_pair_params.pkl'.format(note))
# if ttest.statistic > 0 and ttest.pvalue < 0.05:
# joblib.dump(pair_params, result_dir+'/big_ori/{}_pair_params.pkl'.format(note))
# elif ttest.statistic > 0 and ttest.pvalue >= 0.05:
# joblib.dump(pair_params, result_dir + '/big_ori_non_significant/{}_pair_params.pkl'.format(note))
# elif ttest.statistic <= 0 and ttest.pvalue >= 0.05:
# joblib.dump(pair_params, result_dir + '/big_ot_non_significant/{}_pair_params.pkl'.format(note))
# else:
# joblib.dump(pair_params, result_dir + '/big_ot/{}_pair_params.pkl'.format(note))
print('-----------------------------------------------------------------------------------------')
# print(scores_params)
# note = '{}s_{}ts_{}_{}_{}_sinkhorn_{}_{}'.format(i, len(target_ids), experiment,
# nor_method, clf_method, op_function, metric_xst)
# joblib.dump(scores_params, result_dir + '/{}_score_params.pkl'.format(note))
# print('unsuccessful transport', unsec)
print(time.strftime('%Y-%m-%d %A %X %Z', time.localtime(time.time())))
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
# args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2'
metric_xst = "h"
experiment = 'newfmri'
nor_method = 'no'
clf_method = 'svm'
source_id = 1
target_id = 2
op_function = 'l1l2' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
elif experiment == 'meg':
source = meg
else:
source = newfmri
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
nb_subs = len(indices_sub)
pca = PCA(n_components=144)
i = source_id
j = target_id
xs = x[indices_sub[i]]
xs_pca = pca.fit_transform(xs)
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
xt_pca = pca.fit_transform(xt)
yt = y_target[indices_sub[j]]
note = 'new_pca_h_{}s_{}t_{}_{}_{}_{}'.format(i, j, experiment, nor_method,
clf_method, op_function)
print(note)
# cityblock sqeuclidian, minkowski
best_xst, ot_params, params_acc = fucs.pairsubs_sinkhorn_lables_h(
xs, ys, xt, ot_method=op_function, clf_method=clf_method)
# train on xst, test on xt
ot_accs = fucs.get_accuracy(best_xst, ys, xt, yt, clf_method=clf_method)
ot_score = np.mean(ot_accs)
print('ot source data predicts on original target, accuracy', ot_score,
ot_accs)
# train on xs, test on xt
ori_accs = fucs.get_accuracy(xs, ys, xt, yt, clf_method=clf_method)
ori_score = np.mean(ori_accs)
print('original source data predicts on original target, accuracy',
ori_score, ori_accs)
# train on xs indi-nor, test on xt indi-nor
x_base = x_indi.copy()
xs_base = x_base[indices_sub[i]]
ys_base = y_target[indices_sub[i]]
xt_base = x_base[indices_sub[j]]
yt_base = y_target[indices_sub[j]]
base_accs = fucs.get_accuracy(xs_base,
ys_base,
xt_base,
yt_base,
clf_method=clf_method)
base_score = np.mean(base_accs)
print(
'base: indi-normalized source predicts on indi-normalized target, accuracy',
base_score, base_accs)
pair_params = {}
pair_params['source'] = i
pair_params['target'] = j
pair_params['ori_score'] = round(ori_score, 4)
pair_params['ori_accs'] = ori_accs
pair_params['ot_accs'] = ot_accs
pair_params['ot_score'] = round(ot_score, 4)
pair_params['base_accs'] = base_accs
pair_params['base_score'] = round(base_score, 4)
pair_params['params'] = ot_params
pair_params['params_acc'] = params_acc
ttest_base_ot = stats.ttest_rel(base_accs, ot_accs)
pair_params['ttest'] = (ttest_base_ot.pvalue, ttest_base_ot.statistic)
print('pair_params', pair_params)
ttest_ori_ot = stats.ttest_rel(ori_accs, ot_accs)
pair_params['ttest_ori_ot'] = (ttest_ori_ot.pvalue, ttest_ori_ot.statistic)
if ttest_base_ot.statistic > 0:
joblib.dump(pair_params,
result_dir + '/big_base/{}_pair_params.pkl'.format(note))
else:
joblib.dump(pair_params,
result_dir + '/small_base/{}_pair_params.pkl'.format(note))
if ttest_ori_ot.statistic > 0:
joblib.dump(pair_params,
result_dir + '/big_ori/{}_pair_params.pkl'.format(note))
else:
joblib.dump(pair_params,
result_dir + '/small_ori/{}_pair_params.pkl'.format(note))
reg_coor = np.log10(params_acc['params'])
zero_coor = reg_coor[params_acc['acc'] == 0]
other_coor = reg_coor[params_acc['acc'] != 0]
plt.figure(figsize=(9, 5))
plt.subplot(1, 2, 1)
plt.plot(zero_coor[:, 0], zero_coor[:, 1], '+r', label='acc is 0.5')
plt.plot(other_coor[:, 0], other_coor[:, 1], 'ob', label='acc not 0.5')
plt.legend(loc=0)
plt.title(note)
plt.subplot(1, 2, 2)
accs = params_acc['acc'].reshape((6, 6))
accs2 = np.empty((accs.T.shape))
for k in range(accs.shape[0]):
accs2[:, k] = accs[k][::-1]
plt.imshow(accs2)
plt.colorbar()
plt.title('imshow')
plt.savefig(main_dir + '/figs/acc_imshow_{}.png'.format(note))
print(
'-----------------------------------------------------------------------------------------'
)
# print(scores_params)
# note = '{}s_{}ts_{}_{}_{}_sinkhorn_{}_{}'.format(i, len(target_ids), experiment,
# nor_method, clf_method, op_function, metric_xst)
# joblib.dump(scores_params, result_dir + '/{}_score_params.pkl'.format(note))
# print('unsuccessful transport', unsec)
print(time.strftime('%Y-%m-%d %A %X %Z', time.localtime(time.time())))
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
experiment = 'fmril'
nor_method = 'indi'
clf_method = 'svm'
op_function = 'maplin' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
sub_ids = np.arange(5)
xs1 = [x_data[indices_sub[i]][:20] for i in sub_ids]
xs2 = [x_data[indices_sub[i]][20:] for i in sub_ids]
xs_indi1 = [x_indi[indices_sub[i]][:20] for i in sub_ids]
xs_indi2 = [x_indi[indices_sub[i]][20:] for i in sub_ids]
def barycenter(Xs, loss_fun, reg, max_iter, tol):
S = len(Xs)
Xs = [np.asarray(Xs[s], dtype=np.float64) for s in range(S)]
ps = [ot.utils.unif(Xs[s].shape[0]) for s in range(S)]
p = ot.utils.unif(Xs[0].shape[0])
Cs = [ot.utils.dist(Xs[s], Xs[s], metric='euclidean') for s in range(S)]
Cs = [cs / np.max(cs) for cs in Cs]
weights = ot.utils.unif(S)
C = ot.gromov.gromov_barycenters(Xs[0].shape[0], Cs,
ps, p, weights, loss_fun, reg,
max_iter=max_iter, tol=tol, log=False)
return C
reg = 0.1
C1 = barycenter(xs_indi1, 'square_loss', reg, 100, 1e-5)
C2 = barycenter(xs_indi2, 'square_loss', reg, 100, 1e-5)
xt_indi = x_indi[indices_sub[8]]
p1 = ot.utils.unif(xs_indi1[0].shape[0])
p_t = ot.utils.unif(xs_indi1[0].shape[0])
data1 = xt_indi[:20]
data2 = xt_indi[20:]
c_t1 = ot.utils.dist(data1, data1, metric='euclidean')
eta = 1
gw_distance1 = ot.gromov.gromov_wasserstein2(c_t1, C1, p_t, p1, 'square_loss', eta, 100, 1e-5)
gw_distance2 = ot.gromov.gromov_wasserstein2(c_t1, C2, p_t, p1, 'square_loss', eta, 100, 1e-5)
print(gw_distance1, gw_distance2)
c_t2 = ot.utils.dist(data2, data2, metric='euclidean')
gw_distance1 = ot.gromov.gromov_wasserstein2(c_t2, C1, p_t, p1, 'square_loss', eta, 100, 1e-5)
gw_distance2 = ot.gromov.gromov_wasserstein2(c_t2, C2, p_t, p1, 'square_loss', eta, 100, 1e-5)
print(gw_distance1, gw_distance2)
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi.copy() if nor_method == 'indi' else x_data.copy()
indices_sub = fucs.split_subjects(subjects)
nb_subs = len(indices_sub)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
N = 160
A = fmril.x_data[:N]
# A2 = np.empty(N,60,90)
# for i in range(len(A)):
# A2[i] = A[i].reshape(60,90)
B = np.empty(A.shape)
for i in range(A.shape[0]):
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
# args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2'
metric_xst = 'null'
experiment = 'fmril'
nor_method = 'no'
clf_method = 'logis'
source_id = 1
target_id = 2
op_function = 'l1l2' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi.copy() if nor_method == 'indi' else x_data.copy()
indices_sub = fucs.split_subjects(subjects)
nb_subs = len(indices_sub)
i = source_id
j = target_id
xs = x[indices_sub[i]]
ys = y_target[indices_sub[i]]
xt = x[indices_sub[j]]
# np.savez(result_dir +'/1s_2t_{}_{}_{}.npz'.format(experiment,nor_method,clf_method),xs=xs,xt=xt)
yt = y_target[indices_sub[j]]
note = '{}s_{}t_{}_{}_{}_sinkhorn_{}_{}'.format(i, j, experiment,
nor_method, clf_method, op_function,metric_xst)
print(note)
reg = 0.1
eta = 0.1
print('reg,eta',reg,eta)
xsts = []
gammas = []
for reg in [1e-2, 1e-1, 1, 10]:
for eta in [1e-3,1e-2, 1e-1, 1,10]:
transport = ot.da.SinkhornL1l2Transport
trans_fuc = transport(reg_e=reg, reg_cl=eta, norm='max')
trans_fuc.fit(Xs=xs, Xt=xt, ys=ys)
xst = trans_fuc.transform(Xs=xs)
xsts.append(xst)
np.save(result_dir +'/different_xst.npy',xsts)
def main():
main_dir = os.path.split(os.getcwd())[0]
result_dir = main_dir + '/results'
args = sys.argv[1:]
# experiment = args[0]
# nor_method = args[1]
# clf_method = args[2]
# source_id = int(args[3])
# target_id = int(args[4])
# op_function = args[5] # 'lpl1' 'l1l2' 'maplin'
experiment = 'fmril'
nor_method = 'indi'
clf_method = 'svm'
op_function = 'maplin' # 'lpl1' 'l1l2'
if experiment == 'fmril':
source = fmril
elif experiment == 'fmrir':
source = fmrir
else:
source = meg
result_dir = result_dir + '/{}'.format(experiment)
y_target = source.y_target
subjects = np.array(source.subjects)
x_data = source.x_data_pow if experiment == 'meg' else source.x_data
x_indi = source.x_indi_pow if experiment == 'meg' else source.x_indi
x = x_indi if nor_method == 'indi' else x_data
indices_sub = fucs.split_subjects(subjects)
# j = target_id
# xs = [x[indices_sub[i]] for i in source_id]
# ys = [y_target[indices_sub[i]] for i in source_id]
# xs = np.vstack(xs)
# ys = np.concatenate(ys)
# xt = x[indices_sub[j]]
# yt = y_target[indices_sub[j]]
# xsxt = xs.append(xt)
sub_ids = np.arange(3)
xsxt = [x_data[indices_sub[i]] for i in sub_ids]
xsxt_indi = [x_indi[indices_sub[i]] for i in sub_ids]
ysyt = [y_target[indices_sub[i]] for i in sub_ids]
trans_xsxt = fucs.inverse_xc_barycenters(xsxt_indi,
'square_loss',
0.1,
max_iter=1000,
tol=1e-9)
# trans_xsxt = [StandardScaler().fit_transform(trans_xsxt[i]) for i in sub_ids]
def split_data(data, i, ids):
data = np.asarray(data)
test = data[i]
train = [data[j] for j in ids if j != i]
train = np.vstack(train) if len(
train[0].shape) > 1 else np.concatenate(train)
return train, test
accs = {'trans_accs': [], 'ori_accs': [], 'indi_accs': []}
for i in sub_ids:
trans_xs, trans_xt = split_data(trans_xsxt, i, sub_ids)
ori_xs, ori_xt = split_data(xsxt, i, sub_ids)
indi_xs, indi_xt = split_data(xsxt_indi, i, sub_ids)
ys, yt = split_data(ysyt, i, sub_ids)
_, _, trans_acc, _ = fucs.gridsearch_persub(trans_xs,
ys,
trans_xt,
yt,
clf_method='svm',
njobs=3)
_, _, ori_acc, _ = fucs.gridsearch_persub(ori_xs,
ys,
ori_xt,
yt,
clf_method='svm',
njobs=3)
_, _, indi_acc, _ = fucs.gridsearch_persub(indi_xs,
ys,
indi_xt,
yt,
clf_method='svm',
njobs=3)
accs['trans_accs'].append((i, trans_acc))
accs['ori_accs'].append((i, ori_acc))
accs['indi_accs'].append((i, indi_acc))
print('xt_id, trans_acc, ori_acc, indi_acc', i, trans_acc, ori_acc,
indi_acc)
print(accs)
print(time.strftime('%Y-%m-%d %A %X %Z', time.localtime(time.time())))
import numpy as np
import time
import scipy as sp
import scipy.ndimage as spi
from scipy.spatial.distance import cdist
import matplotlib.pylab as pl
from sklearn import manifold
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.metrics.pairwise import rbf_kernel
import ot
x = fmril.x_indi
ids = fucs.split_subjects(fmril.subjects)
for source in range(5):
for target in range(5):
if source != target:
data1 = x[ids[source]]
y1 = fmril.y_target[ids[source]]
data2 = x[ids[target]]
y2 = fmril.y_target[ids[target]]
C = ot.utils.dist(data1, data1)
max_iter = 3000
eps = 1e-9
dim = 5400