def plot_lc_all():
# Load test set learning curve values
scores = {}
models = ['linear', 'ridge', 'lasso', 'xgb_a']
colors = ['green', 'c', 'orangered', 'mediumslateblue', 'saddlebrown']
for i in range(0, len(models)):
filepath = get_abspath('{0}/{0}_LC_test.csv'.format(models[i]),
'outputs')
scores[models[i]] = (np.mean(pd.read_csv(filepath), axis=1), colors[i])
# Set training sizes and intervals
train_sizes = np.arange(0.01, 1.0, 0.05)
# Create learning curve plot
fig, ax = plt.subplots()
for k, v in scores.items():
ax.plot(train_sizes, v[0], marker='.', color=v[1], label=k)
ax.legend(loc='best')
ax.grid(linestyle='dotted')
ax.set_title('Learning curves - test score comparison')
ax.set_xlabel('Samples used for training as a percentage of total')
ax.set_ylabel('RMSE (Negative)')
# Save learning curve plot as PNG
plotpath = get_abspath('models_LC.png', 'graphs')
ensure_dir_exists(plotpath)
plt.tight_layout(pad=1.0)
plt.savefig(plotpath)
plt.close()
def main():
config = db_config()
conn = db_conn(config)
# Create scorer to train models using RMSE
scorer = make_scorer(rmse, greater_is_better=False)
# Load features
features_a = preprocess_features(conn)
features_b = preprocess_features(conn, with_sa=False)
# Get train and test sets for both feature sets
X_train_a, X_test_a, y_train_a, y_test_a = split_data(features_a)
X_train_b, X_test_b, y_train_b, y_test_b = split_data(features_b)
# Train regression models with sentiment score features
modelpath = 'models'
linear_models = {
'linear': LinearRegression(),
'ridge': Ridge(),
'lasso': Lasso()
}
for name, model in linear_models.items():
grid = train_model(model_object=model,
model_type=name,
X_train=X_train_a,
y_train=y_train_a,
score_function=scorer,
cv=3)
filename = get_abspath('{}.model'.format(name), modelpath)
ensure_dir_exists(filename)
save_pickled_model(grid, filename)
# Train XGBoost model with sentiment score features
xgb = XGBRegressor(objective='reg:linear')
grid = train_model(model_object=xgb,
model_type='xgb',
X_train=X_train_a,
y_train=y_train_a,
score_function=scorer,
cv=3)
filename = get_abspath('xgb_a.model', modelpath)
save_pickled_model(grid, filename)
# Train XGBoost model without sentiment score features
grid = train_model(model_object=xgb,
model_type='xgb',
X_train=X_train_b,
y_train=y_train_b,
score_function=scorer,
cv=3)
filename = get_abspath('xgb_b.model', modelpath)
save_pickled_model(grid, filename)
def backup(args):
db = args.db
out = get_abspath(args.output_dir)
now = datetime.datetime.now()
out = os.path.join(out, '%s__%s' % (db, now.strftime('%Y_%m_%d_%H%M%S')))
cmd = ['mongodump', '-o', '%s' % out]
execute(cmd, args)
def plot_learning_curve(model_name, train_scores, test_scores):
# Set training sizes and intervals
train_sizes = np.arange(0.01, 1.0, 0.05)
# Create learning curve plot
fig, ax = plt.subplots()
ax.plot(train_sizes,
np.mean(train_scores, axis=1),
marker='.',
color='b',
label='Training score')
ax.plot(train_sizes,
np.mean(test_scores, axis=1),
marker='.',
color='g',
label='Cross-validation score')
ax.legend(loc='best')
ax.grid(linestyle='dotted')
ax.set_title('Learning curve - {} model'.format(model_name))
ax.set_xlabel('Samples used for training as a percentage of total')
ax.set_ylabel('RMSE (Negative)')
# Save learning curve plot as PNG
plot_tgt = '{0}/{1}'.format('graphs', model_name)
plotpath = get_abspath('{}_LC.png'.format(model_name), plot_tgt)
ensure_dir_exists(plotpath)
plt.tight_layout(pad=1.0)
plt.savefig(plotpath)
plt.close()
def create_learning_curve(estimator,
scorer,
X_train,
y_train,
model_name,
cv=3,
save_file=True):
# Set training sizes and intervals
train_sizes = np.arange(0.01, 1.0, 0.05)
# Set cross validation strategy to use StratifiedShuffleSplit
cv_strategy = StratifiedShuffleSplit(n_splits=cv, random_state=0)
# Create learning curve object
LC = learning_curve(estimator,
X_train,
y_train,
cv=cv_strategy,
train_sizes=train_sizes,
scoring=scorer,
n_jobs=-1)
# Extract training and test scores as data frames
train_scores = pd.DataFrame(index=LC[0], data=LC[1])
test_scores = pd.DataFrame(index=LC[0], data=LC[2])
# Save data frames to CSV
if save_file:
res_tgt = '{0}/{1}'.format('outputs', model_name)
train_file = get_abspath('{}_LC_train.csv'.format(model_name), res_tgt)
test_file = get_abspath('{}_LC_test.csv'.format(model_name), res_tgt)
ensure_dir_exists(train_file)
train_scores.to_csv(train_file, index=False)
test_scores.to_csv(test_file, index=False)
return train_scores, test_scores
def plot_score_histograms(df, score_type, filename, filepath='graphs'):
# Set figure parameters
sns.set(font_scale=1.3, rc={'figure.figsize': (12, 8)})
# Create plot and set parameters
fig, ax = plt.subplots()
ax.hist(df, bins=20)
fig.suptitle('{} Polarity Score Frequency'.format(score_type))
ax.set_xlabel('Value')
ax.set_ylabel('Frequency')
# Save plot
plotpath = get_abspath(filename, filepath)
ensure_dir_exists(plotpath)
plt.savefig(plotpath)
plt.close()
def basic_results(grid, X_test, y_test, model_name):
# Get best score, test score, scoring function and best parameters
bs = grid.best_score_
ts = grid.score(X_test, y_test)
sf = grid.scorer_
bp = grid.best_params_
# Calculate R^2
y_pred = grid.predict(X_test)
r2 = r2_score(y_test, y_pred)
# Write results to a combined results file
resfile = get_abspath('basic_results.csv', 'outputs')
ensure_dir_exists(resfile)
with open(resfile, 'a') as f:
f.write('{0},{1},{2},"{3}","{4}",{5}\n'.format(model_name, bs, ts, sf,
bp, r2))
def get_feature_importances(model_name, grid, features, save_file=True):
# Get feature importance values
feat_importance = pd.Series(
grid.best_estimator_.get_booster().get_fscore()).sort_values(
ascending=False)
# Get feature names
feat_idxs = [i.strip('f') for i in feat_importance.keys()]
feat_names = list(features.columns.values)
feats = [x for _, x in sorted(zip(feat_idxs, feat_names))]
# Add feature names
df = feat_importance.to_frame(name='f_score')
df.reset_index(drop=True, inplace=True)
df['name'] = feats
# Save feature importance to CSV
if save_file:
file_tgt = '{0}/{1}'.format('outputs', model_name)
feats_file = get_abspath('{}_FI.csv'.format(model_name), file_tgt)
df.to_csv(feats_file, index=False)
return df
def create_validation_curve(estimator, X_train, y_train, model_name,
param_name, param_range, scorer):
# Generate validation curve results
train_scores, test_scores = validation_curve(estimator,
X_train,
y_train,
param_name=param_name,
param_range=param_range,
cv=3,
scoring=scorer,
n_jobs=-1)
# Generate validation curve plot
fig, ax = plt.subplots()
ax.plot(param_range,
np.mean(train_scores, axis=1),
marker='.',
color='b',
label='Train Score')
ax.plot(param_range,
np.mean(test_scores, axis=1),
marker='.',
color='g',
label='Cross-validation Score')
ax.legend(loc='best')
ax.grid(linestyle='dotted')
ax.set_title('Validation curve - {} model'.format(model_name))
ax.set_xlabel(param_name)
ax.set_ylabel('RMSE (Negative)')
# Save validation curve plot as PNG
plot_tgt = '{0}/{1}'.format('graphs', model_name)
plotpath = get_abspath('{}_VC.png'.format(model_name), plot_tgt)
ensure_dir_exists(plotpath)
plt.tight_layout(pad=1.0)
plt.savefig(plotpath)
plt.close()
def plot_feature_importances(model_name, feature_importances, nfeats=10):
# Subset feature importances data frame
subset = feature_importances.iloc[:nfeats, :]
# Generate feature importance plot
fig, ax = plt.subplots()
sns.barplot(x='f_score',
y='name',
data=subset,
ax=ax,
palette=sns.color_palette('YlGnBu_r', n_colors=15),
dodge=False)
ax.tick_params(labelsize=8)
ax.set_ylabel('')
ax.set_xlabel('F score')
ax.set_title('Feature importances - {} model'.format(model_name))
# Save feature importance plot as PNG
plot_tgt = '{0}/{1}'.format('graphs', model_name)
plotpath = get_abspath('{}_FI.png'.format(model_name), plot_tgt)
ensure_dir_exists(plotpath)
plt.tight_layout(pad=1.0)
plt.savefig(plotpath)
plt.close()
parser = argparse.ArgumentParser(
prog='detect',
description='Detect faces in an image.'
)
parser.add_argument(
'path',
type=str,
help='path or URL of a photo where faces will be detected')
args = parser.parse_args()
# ----------------------------------------------------------------------
# Setup
# ----------------------------------------------------------------------
img_url = args.path if is_url(args.path) else get_abspath(args.path)
face_attrs = ['age', 'gender', 'glasses', 'emotion', 'occlusion']
# ----------------------------------------------------------------------
# Call face API to detect and describe faces
# ----------------------------------------------------------------------
# Request subscription key and endpoint from user.
subscription_key, endpoint = get_private()
credentials = CognitiveServicesCredentials(subscription_key) # Set credentials
client = FaceClient(endpoint, credentials) # Setup Azure face API client
faces = azface_detect(client, img_url, return_face_attributes=face_attrs)
print_detection_results(faces)
# ----------------------------------------------------------------------
parser = argparse.ArgumentParser(prog='score',
parents=[option_parser],
description='Detect faces in an image.')
parser.add_argument('image', type=str, help='image path or URL')
args = parser.parse_args()
# Wrap face detection parameters.
face_params = FaceParams(args.scaleFactor, args.minNeighbors, args.minSize)
# ----------------------------------------------------------------------
# Face detection
# ----------------------------------------------------------------------
image = read_cv_image_from(
args.image if is_url(args.image) else get_abspath(args.image))
faces = detect_faces(image, face_params=face_params)
print("Found {0} faces!".format(len(faces)))
result = mark_faces(image, faces)
image, result = convert_cv2matplot(image, result)
plot_side_by_side_comparison(image, result, rightlabel="Detected Faces")
)
parser.add_argument(
'target',
help='path or URL of a photo of the faces to be found')
parser.add_argument(
'candidate',
help='path or URL of a photo to find expected target faces')
args = parser.parse_args()
# ----------------------------------------------------------------------
# Setup
# ----------------------------------------------------------------------
target_url = args.target if is_url(args.target) else get_abspath(args.target) # Get the photo of target faces
candidate_url = args.candidate if is_url(args.candidate) else get_abspath(args.candidate) # Get the photo to be checked
if os.path.isdir(target_url) or os.path.isdir(candidate_url):
stop("Only one photo allowed!")
# ----------------------------------------------------------------------
# Prepare Face API client
# ----------------------------------------------------------------------
# Request subscription key and endpoint from user.
subscription_key, endpoint = get_private()
credentials = CognitiveServicesCredentials(subscription_key) # Set credentials
client = FaceClient(endpoint, credentials) # Setup Azure face API client
def main():
config = db_config()
conn = db_conn(config)
# Remove basic results file
try:
combined = get_abspath('basic_results.csv', 'outputs')
os.remove(combined)
except IOError:
pass
# Load features
features_a = preprocess_features(conn)
features_b = preprocess_features(conn, with_sa=False)
# Create scorer to train models using RMSE
scorer = make_scorer(rmse, greater_is_better=False)
# Load models in a dict
models = {
'linear': load_pickled_model('models/linear.model'),
'ridge': load_pickled_model('models/ridge.model'),
'lasso': load_pickled_model('models/lasso.model'),
'xgb_a': load_pickled_model('models/xgb_a.model'),
'xgb_b': load_pickled_model('models/xgb_b.model')
}
# Validation curve parameter names and ranges
vc_params = {
'xgb_a': ('max_depth', np.arange(1, 20, 1)),
'xgb_b': ('max_depth', np.arange(1, 20, 1))
}
# Split into train and test sets
X_train_a, X_test_a, y_train_a, y_test_a = split_data(features_a)
X_train_b, X_test_b, y_train_b, y_test_b = split_data(features_b)
# Generate basic results and learning curves for all models
for name, grid in models.items():
if name in ['linear', 'ridge', 'lasso']:
basic_results(grid, X_test_a, y_test_a, name)
train_scores, test_scores = create_learning_curve(
grid.best_estimator_,
scorer,
X_train_a,
y_train_a,
model_name=name,
cv=3)
plot_learning_curve(name, train_scores, test_scores)
if name == 'xgb_b':
basic_results(grid, X_test_b, y_test_b, name)
train_scores, test_scores = create_learning_curve(
grid.best_estimator_,
scorer,
X_train_b,
y_train_b,
model_name=name,
cv=3)
plot_learning_curve(name, train_scores, test_scores)
# Generate validation curves for XGBoost models
create_validation_curve(models['xgb_a'].best_estimator_,
X_train_a,
y_train_a,
model_name='xgb_a',
param_name=vc_params['xgb_a'][0],
param_range=vc_params['xgb_a'][1],
scorer=scorer)
create_validation_curve(models['xgb_b'].best_estimator_,
X_train_b,
y_train_b,
model_name='xgb_b',
param_name=vc_params['xgb_b'][0],
param_range=vc_params['xgb_b'][1],
scorer=scorer)
# Generate XGBoost feature importance plots and results
fi_a = get_feature_importances('xgb_a', models['xgb_a'], features_a)
fi_b = get_feature_importances('xgb_b', models['xgb_b'], features_b)
plot_feature_importances('xgb_a', fi_a, nfeats=15)
plot_feature_importances('xgb_b', fi_b, nfeats=15)
# Plot test set learning curves of all five models
plot_lc_all()
def restore(args):
input_dir = get_abspath(args.input_dir)
cmd = ['mongorestore', '%s' % input_dir]
execute(cmd, args)