alpha = 0.3
USE_CACHE = False
DUMP_CACHE = False
cv = CrossValidationManager('RET')
configurations = [
'{}{}'.format(cv.get_run_id(i)[0],
cv.get_run_id(i)[1])
for i in ([0, 1, 3, 4, 7] if dset == 'Valid' else [2, 5, 6, 8, 9])
] # [range(10)]
#configurations = range(n_groups)
# configurations = [1]
dset = 'Valid'
runs, run_net_types, run_metrics, run_epochs, run_names, _, _ = load_experiments(
experiment_name)
for m, metric_rating in enumerate(rating_metrics):
Valid_epochs, Idx_malig_pearson, Idx_malig_kendall, Idx_rating_pearson, Idx_rating_kendall = [], [], [], [], []
for run, net_type, dist, epochs, metric in zip(runs, run_net_types,
run_metrics, run_epochs,
run_metrics):
plot_data_filename = './Plots/Data/correlation_{}{}.p'.format(
net_type, run)
try:
if USE_CACHE is False:
print('NOTE: SKIPPING TO EVELUATION')
assert False
valid_epochs, idx_malig_pearson, idx_malig_kendall, idx_rating_pearson, idx_rating_kendall = \
if __name__ == "__main__":
# Setup
exp_name = 'DirSimilarityLoss'
dset = 'Test'
rating_norm = 'none'
n_groups = 5
start = timer()
num_of_indexes = 3 + 4
# cv = CrossValidationManager('RET')
# configurations = ['{}{}'.format(cv.get_run_id(i)[0], cv.get_run_id(i)[1]) for i in [0, 1, 3, 4, 7]] # [range(10)]
configurations = range(n_groups)
# configurations = [1]
runs, run_net_types, run_metrics, _, run_names, run_ep_perf, run_ep_comb = load_experiments(
exp_name)
data = np.zeros((len(runs), num_of_indexes))
dataStd = np.zeros((len(runs), num_of_indexes))
# evaluate
run_id = 0
for run, net_type, _, metric, epochs in zip(runs, run_net_types,
range(len(runs)), run_metrics,
run_ep_perf):
print("Evaluating classification accuracy for {}{}".format(
net_type, run))
acc, acc_std, _ = eval_classification(run=run,
net_type=net_type,
dset=dset,
metric=metric,
def plot_row(i, distances, label=None):
tau, l_e = index.kumar(distances, res=0.01)
conc = index.concentration(distances)
contrast = index.relative_contrast_imp(distances)
p[i].hist(distances.flatten() / np.mean(distances), bins=20)
p[i].axes.title.set_text('std:{:.2f}, conc:{:.2f}, ctrst:{:.2f}'.format(
conc[2], conc[0], contrast[0]))
p[i].axes.yaxis.label.set_text('distribution')
p[i].axes.xaxis.label.set_text('distances')
Ret.pca(epoch=e, plt_=p[i + 1], label=label)
p[i + 2].plot(l_e[1], l_e[0])
p[i + 2].axes.title.set_text('Kumari (tau = {:.2f}'.format(tau))
runs, run_net_types, run_metrics, run_epochs, run_names, _, _ = load_experiments(
'SiamRating')
# evaluate
Epochs, Idx_hubness, Idx_symmetry, Idx_concentration, Idx_contrast, Idx_kummar = [], [], [], [], [], []
for m, metric in enumerate(metrics):
print("Begin: {} metric".format(metric))
for run, net_type, r, epochs, name in zip(runs, run_net_types,
range(len(runs)), run_epochs,
run_names):
print("Evaluating run {}{}".format(net_type, run))
# initialize figures
plt.figure("Distances - {}".format(name))
p = [None] * 9
for i in range(9):