target_v = Variable(y_test.cuda(cuda_id), requires_grad=False)
input_train = x_train
target_train = y_train
n_features = input_v.size(1)
trainer = TrainerSimpleNet(n_classes, n_features, n_epochs, batch_size, cuda_id)
trainer.fit(input_train, target_train)
res = trainer.evaluate(input_v, target_v)
print('Test: loss={:.4}, acc={:.2f}%, f1={:.4}'.format(*res))
return (fold_id, res)
#%%
if __name__ == "__main__":
feat_data, col2ignore_r = read_feats()
#%%
# core_sets = [
# ['length_50th',
# 'curvature_head_IQR',
# 'quirkiness_50th',
# 'speed_90th',
# 'motion_mode_backward_fraction'],
#
# ['curvature_head_abs_90th',
# 'width_midbody_norm_50th',
# 'length_50th',
# 'curvature_hips_abs_90th',
# 'motion_mode_backward_fraction'],
print(db_name, x_t, yy[min_ind], yy.max())
plt.title(db_name)
ff = os.path.join(save_dir, '{}_{}.pdf'.format(experimental_dataset,
db_name))
plt.savefig(ff)
#%%
#%%
#I forgot to add the last feature remaining so I have to do a dirty hack
if os.path.basename(save_name).startswith('R_'):
if True:
#i forgot to add the last feature extracted...
feat_data, col2ignore_r = read_feats(experimental_dataset)
all_feats = [
x for x in feat_data['tierpsy'].columns
if x not in col2ignore_r
]
del feat_data
#remove ventral signed columns that where not abs (This ones seemed useless...)
v_cols = [
x for x in all_feats if not (('eigen' in x) or ('blob' in x))
]
v_cols_remove = [
x.replace('_abs', '') for x in v_cols if '_abs' in x
]
all_feats = list(set(v_cols) - set(v_cols_remove))
all_feats = set(all_feats)
top16_manual = [
'length_90th', 'width_midbody_norm_10th', 'curvature_hips_abs_90th',
'curvature_head_abs_90th', 'motion_mode_paused_fraction',
'motion_mode_paused_frequency', 'd_curvature_hips_abs_90th',
'd_curvature_head_abs_90th', 'width_head_base_norm_10th',
'motion_mode_backward_frequency', 'quirkiness_50th', 'minor_axis_50th',
'curvature_midbody_norm_abs_50th',
'relative_to_hips_radial_velocity_tail_tip_50th',
'relative_to_head_base_radial_velocity_head_tip_50th',
'relative_to_head_base_angular_velocity_head_tip_abs_90th'
]
if __name__ == '__main__':
experimental_dataset = 'SWDB'
feat_data, col2ignore_r = read_feats(experimental_dataset,
z_transform=False)
#feat_data, col2ignore_r = read_feats(experimental_dataset, z_transform = True)
df = feat_data['tierpsy']
del feat_data
#this videos are clearly outliers. They are from 2009. Andre's normally says this date was
#before they developed the final experimental protocol so I assume therewas a problem with them.
bad_index = df.index[(df['strain'] == 'N2') & (df['length_50th'] > 1500)]
df.drop(bad_index, inplace=True)
#%%
strain_sets = dict(mutants=[
'Schafer Lab N2 (Bristol, UK)', 'dpy-20(e1282)IV', 'egl-5(n486)III',
'unc-9(e101)X', 'unc-77(e625)IV', 'ser-4(ok512)III', 'sma-2(e502)III',
'trp-4(sy695)I'