def linear_regression_single(record_predictions=False, record_loss=True):
labels, data, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_luminance=True)
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_luminance=True)
traits = ['O', 'C', 'E', 'A', 'S']
for i, t in enumerate(traits):
print('for trait: %s' % t)
label_trait = labels[:, i]
reg = LinearRegression().fit(data, label_trait)
print(
'reg score: ', reg.score(data, label_trait)
) # 0.011438774006662768, 0.01160391566097031, 0.011790330654130819
prediction = reg.predict(data_test)
loss = np.abs(prediction - labels_test[:, i])
print('loss: ', np.mean(loss))
if record_loss:
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE,
'testall_%s_%s.txt' % (num, t))
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
if record_predictions:
U.record_all_predictions(which='test', preds=prediction, trait=t)
def linear_regression_single():
labels, data, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_color=True)
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_color=True)
traits = ['O', 'C', 'E', 'A', 'S']
for i, t in enumerate(traits):
print('for trait: %s' % t)
label_trait = labels[:, i]
reg = LinearRegression().fit(data, label_trait)
print('reg score: ', reg.score(data, label_trait))
prediction = reg.predict(data_test)
loss = np.abs(prediction - labels_test[:, i])
print('loss: ', np.mean(loss))
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE, 'testall_%s_%s.txt' % (num, t))
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
def linear_regression_all(record_predictions=False, record_loss=True):
labels, data, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_luminance=True)
reg = LinearRegression().fit(data, labels)
reg.score(data, labels)
print('reg score: ', reg.score(data, labels)
) # 0.011438774006662768, 0.01160391566097031, 0.011790330654130819
labels, data, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_luminance=True)
prediction = reg.predict(data)
loss = np.abs(prediction - labels)
loss_copy = loss
loss = np.mean(loss, axis=1)
# loss = np.mean(loss)
print('loss: ', np.mean(loss))
print('OCEAS: ', np.mean(loss_copy, axis=0))
if record_loss:
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE, 'testall_%s.txt' % num)
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
if record_predictions:
U.record_all_predictions(which='test', preds=prediction)
def linear_regression_all():
labels, data, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_color=True)
reg = LinearRegression().fit(data, labels)
reg.score(data, labels)
print('reg score: ', reg.score(data,
labels)) # reg score: 0.012926151576513799
labels, data, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_color=True)
prediction = reg.predict(data)
loss = np.abs(prediction - labels)
loss = np.mean(loss, axis=1)
# loss = np.mean(loss)
print('loss: ', np.mean(loss))
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE, 'testall_%s.txt' % num)
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
def ml_lum_clr_geth_all():
# get luminance and color
labels_train, data_train, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_everything=True)
all_labels = D.load_labels_only(list(train_labels), train_labels)
labels = all_labels[:, :5]
# get gender and ethnicity
data = all_labels[:, -2:]
data = np.concatenate(
(data_train, data),
axis=1) # luminance 1, color 3, gender 1, ethnicity 1
reg = LinearRegression().fit(data, labels)
reg.score(data, labels)
print('reg score: ', reg.score(data, labels)) # 0.02931919237388262
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_everything=True)
all_labels = D.load_labels_only(list(test_labels), test_labels)
labels = all_labels[:, :5]
data = all_labels[:, -2:]
data = np.concatenate(
(data_test, data),
axis=1) # luminance 1, color 3, gender 1, ethnicity 1
prediction = reg.predict(data)
loss = np.abs(prediction - labels)
loss = np.mean(loss, axis=1)
print('loss: ', np.mean(loss))
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE, 'testall_%s.txt' % num)
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
def make_trait_lum_plot(num='75'):
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_color=True)
print('data loaded')
def get_color(clr_code):
g = int(clr_code * 100)
# rgb = np.array([255, g, 0])
# rgb = np.expand_dims(rgb, axis=0)
return g
traits = ['O', 'C', 'E', 'A', 'S']
for i, t in enumerate(traits):
labels = labels_test[:, i]
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
x = data_test[:, 0]
y = data_test[:, 1]
z = data_test[:, 2]
col = np.zeros((len(labels)))
for j in range(len(labels)):
col[j] = get_color(labels[j])
ax.scatter(x, y, z, c=col)
# for l in range(len(labels)):
# x, y, z = data_test[l]
# ax.scatter(x, y, z, c=get_color(labels[l]))
ax.set_xlabel('X Label')
ax.set_ylabel('Y Label')
ax.set_zlabel('Z Label')
for angle in range(0, 360, 10):
ax.view_init(30, angle)
plt.draw()
# x = data_test
# y = labels
# plt.scatter(x, y, s=1, alpha=0.8)
# plt.title('trait %s as a function of frame luminance' % t)
# plt.xlabel('frame luminance ')
# plt.ylabel('trait intensity')
save_path = os.path.join(P.FIGURES, 'train_%s' % num)
if not os.path.exists(save_path):
os.mkdir(save_path)
plt.savefig('%s/luminance_angle_%d_%s.png' % (save_path, angle, t))
def make_trait_lum_plot(num='68'):
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_luminance=True)
print('data loaded')
traits = ['O', 'C', 'E', 'A', 'S']
for i, t in enumerate(traits):
labels = labels_test[:, i]
plt.figure()
x = data_test
y = labels
plt.scatter(x, y, s=1, alpha=0.8)
plt.title('trait %s as a function of frame luminance' % t)
plt.xlabel('frame luminance ')
plt.ylabel('trait intensity')
save_path = os.path.join(P.FIGURES, 'train_%s' % num)
if not os.path.exists(save_path):
os.mkdir(save_path)
plt.savefig('%s/luminance_%s.png' % (save_path, t))
def ml_lum_clr_geth_single():
labels_train, data_train, _ = D.load_data_special(list(train_labels),
train_labels,
id_frames,
use_everything=True)
all_labels = D.load_labels_only(list(train_labels), train_labels)
labels = all_labels[:, :5]
# get gender and ethnicity
data = all_labels[:, -2:]
data = np.concatenate(
(data_train, data),
axis=1) # luminance 1, color 3, gender 1, ethnicity 1
labels_test, data_test, _ = D.load_data_special(list(test_labels),
test_labels,
id_frames,
ordered=True,
use_everything=True)
all_labels = D.load_labels_only(list(test_labels), test_labels)
labels_test = all_labels[:, :5]
data_geth = all_labels[:, -2:]
data_test = np.concatenate(
(data_test, data_geth),
axis=1) # luminance 1, color 3, gender 1, ethnicity 1
traits = ['O', 'C', 'E', 'A', 'S']
for i, t in enumerate(traits):
if t == 'A':
print('uo')
print('for trait: %s' % t)
label_trait = labels[:, i]
reg = LinearRegression().fit(data, label_trait)
print('reg score: ', reg.score(data, label_trait))
prediction = reg.predict(data_test)
loss = np.abs(prediction - labels_test[:, i])
print('loss: ', np.mean(loss))
num = P.TEST_LOG.split('_')[-1].split('.')[0]
save_path = os.path.join(P.LOG_BASE, 'testall_%s_%s.txt' % (num, t))
with open(save_path, 'a') as mf:
for i in range(len(loss)):
line = '%s\n' % str(loss[i])[0:6]
mf.write(line)
# for trait: O
# reg score: 0.03884013482567261
# loss: 0.115803376
# for trait: C
# reg score: 0.031371301668408624
# loss: 0.12661433
# for trait: E
# reg score: 0.05366489112270434
# loss: 0.12337612
# uo
# for trait: A
# reg score: 0.011632718380941487
# loss: 0.10779269
# for trait: S
# reg score: 0.013905780240829002
# loss: 0.1263367
# ml_lum_clr_geth_single()