def learning_curve(seed=2, runs=10, light=False, K=2, outlier_threshold=10, granularity=1.0):
sk = Sidekick(data_dir=data_dir, seed=seed)
sk.load(light=light)
rmse_failed_all = []
rmse_success_all = []
rmse_all = []
accuracy_all = []
for r in range(runs):
projects_train, projects_test = sk.split(threshold=0.7, shuffle=True)
n_test = len(projects_test)
projects_validation = projects_test[:floor(n_test*3/5)]
projects_test = projects_test[floor(n_test*3/5):]
_, _, rmse_run, accuracy_run = one_run(projects_train, projects_test, K, outlier_threshold, granularity)
# rmse_failed_all.append(rmse_failed_run)
# rmse_success_all.append(rmse_success_run)
rmse_all.append(rmse_run)
accuracy_all.append(accuracy_run)
# with open('rmse_failed_outlier_%s.pkl' % outlier_threshold, 'wb') as f:
# cp.dump(rmse_failed_all, f)
# with open('rmse_success_outlier_%s.pkl' % outlier_threshold, 'wb') as f:
# cp.dump(rmse_success_all, f)
with open('rmse_K_%s_outlier_%s_granularity_%s.pkl' % (K, outlier_threshold, granularity), 'wb') as f:
cp.dump(rmse_all, f)
with open('accuracy_K_%s_outlier_%s_granularity_%s.pkl' % (K, outlier_threshold, granularity), 'wb') as f:
cp.dump(accuracy_all, f)
return rmse_failed_all, rmse_success_all, rmse_all, accuracy_all
def learning_curve(seed=2,
runs=10,
light=False,
K=2,
outlier_threshold=10,
granularity=1.0):
sk = Sidekick(data_dir=data_dir, seed=seed)
sk.load(light=light)
rmse_failed_all = []
rmse_success_all = []
rmse_all = []
accuracy_all = []
for r in range(runs):
projects_train, projects_test = sk.split(threshold=0.7, shuffle=True)
n_test = len(projects_test)
projects_validation = projects_test[:floor(n_test * 3 / 5)]
projects_test = projects_test[floor(n_test * 3 / 5):]
_, _, rmse_run, accuracy_run = one_run(projects_train, projects_test,
K, outlier_threshold,
granularity)
# rmse_failed_all.append(rmse_failed_run)
# rmse_success_all.append(rmse_success_run)
rmse_all.append(rmse_run)
accuracy_all.append(accuracy_run)
# with open('rmse_failed_outlier_%s.pkl' % outlier_threshold, 'wb') as f:
# cp.dump(rmse_failed_all, f)
# with open('rmse_success_outlier_%s.pkl' % outlier_threshold, 'wb') as f:
# cp.dump(rmse_success_all, f)
with open(
'rmse_K_%s_outlier_%s_granularity_%s.pkl' %
(K, outlier_threshold, granularity), 'wb') as f:
cp.dump(rmse_all, f)
with open(
'accuracy_K_%s_outlier_%s_granularity_%s.pkl' %
(K, outlier_threshold, granularity), 'wb') as f:
cp.dump(accuracy_all, f)
return rmse_failed_all, rmse_success_all, rmse_all, accuracy_all
def learning_curve(seed=2, runs=10, light=False, outlier_threshold=10):
sk = Sidekick(data_dir=data_dir, seed=seed)
sk.load(light=light)
rmse_all = []
accuracy_all = []
for r in range(runs):
projects_train, projects_test = sk.split(threshold=0.7, shuffle=True)
n_test = len(projects_test)
projects_validation = projects_test[:floor(n_test * 3 / 5)]
projects_test = projects_test[floor(n_test * 3 / 5):]
rmse_run, accuracy_run = one_run(projects_train, projects_test, outlier_threshold)
rmse_all.append(rmse_run)
accuracy_all.append(accuracy_run)
with open('gp_rmse_outlier_%s_normalized_false.pkl' % outlier_threshold, 'wb') as f:
cp.dump(rmse_all, f)
with open('gp_accuracy_outlier_%s_normalized_false.pkl' % outlier_threshold, 'wb') as f:
cp.dump(accuracy_all, f)
return rmse_all, accuracy_all
def experiment(args):
"""
Run the experiment for the given number of times.
:param seed: Seed to use when shuffling the data set
:param runs: Number of times to run the experiment
:param light: Whether to use a light data set (1000 projects)
:param outlier_threshold: Threshold of outliers to discard
:param normalized: Whether to use the normalized money
:param granularity: Level of granularity
:return:
"""
features = _get_extractor(args.features)
sk = Sidekick(data_dir=data_dir, seed=args.seed)
sk.load(light=args.light)
relative_time = np.linspace(0.025, 1, 40)
# Construct data dict
data_rmse = {
"plot_label": args.features,
"x": relative_time,
"y": [],
"args": vars(args),
"timestamp": time.time()
}
data_accuracy = {
"plot_label": args.features,
"x": relative_time,
"y": [],
"args": vars(args),
"timestamp": time.time()
}
rmse_all = []
accuracy_all = []
for r in range(args.runs):
projects_train, projects_test = sk.split(threshold=0.7, shuffle=True)
# Set which money time series to use
for p in np.append(projects_train, projects_test):
p.normalized = args.normalized
# n_test = len(projects_test)
# projects_validation = projects_test[:floor(n_test*3/5)]
# projects_test = projects_test[floor(n_test*3/5):]
# Run the experiment once
rmse_run, accuracy_run = _one_run(projects_train, projects_test,
relative_time, features,
args.outlierThreshold,
args.normalized, args.granularity)
# Record the results
rmse_all.append(rmse_run)
accuracy_all.append(accuracy_run)
data_rmse["y"] = rmse_all
data_accuracy["y"] = accuracy_all
# Save the results to disk
args.metric = "rmse"
u.save_args(data_rmse, vars(args))
args.metric = "accuracy"
u.save_args(data_accuracy, vars(args))
return rmse_all, accuracy_all