def main(args):
outdir = args.outdir
# global database
db = s_qs.SpecimenQueries(database_path=args.database)
if not args.userids:
cvd_user_info = get_cvd(db, args.n_users, min_obs=args.min_obs)
cvd_user_info = rank_cvd(cvd_user_info,
alpha=args.alpha).head(args.n_users)
else:
args.userids = args.userids.split(',')
cvd_user_info = compute_cvd_info(db, args.userids)
cvd_user_info = rank_cvd(cvd_user_info, alpha=args.alpha)
# save a copy of the info to a folder for later reference
table_path = os.path.join(outdir, 'cvd_users_info.csv')
cvd_user_info.to_csv(table_path, index=False)
# just the user ids
cvd_user_ids = cvd_user_info['userid'].values
# produce diagnostic plots for each user
if args.plot:
for user in cvd_user_ids:
print("Plotting %s" % user)
ps = plot_cvd_diagnostics(db, user, combined_plots=args.combine)
for p_nm, p in ps.items():
plot_path = os.path.join(
outdir, 'cvd_diagnostics_%s_%s.pdf' % (p_nm, user))
p.savefig(plot_path)
plt.close('all')
print(args.sep.join(map(str, cvd_user_ids)))
def _get_userid(self, n=0):
""" Get userid for the nth user, when sorted by descending number of selections """
db = s_qs.SpecimenQueries(database_path=self.database_path)
cts_by_users = db.execute_adhoc("""
select selectionevents.uniqueId as userid, count(*) as ct
from selectionevents inner join users on selectionevents.userid = users.id
where userid > 1 group by selectionevents.uniqueId
""")
cts_by_users = cts_by_users.sort_values('ct', ascending=False)
return cts_by_users.iloc[n]['userid']
def main(database_path, targetdir, specimendir, scorer, userid, outdir,
target_ncolors):
dfs = [
compute_hue_accuracies(targetdir, specimendir, scorer, userid, ncolors)
for ncolors in target_ncolors
]
db = s_qs.SpecimenQueries(database_path=database_path)
images_df = pd.concat(dfs, axis=0)
observed = get_user_data(db, userid)
observed = observed.groupby(
'target_hue_bucket')['correct'].mean().reset_index()
observed = observed.rename(columns={
'target_hue_bucket': 'hue',
'correct': 'acc'
})
observed['ncolors'] = 'Specimen'
dfs.append(observed)
combined = pd.concat(dfs, axis=0)
print("Correlation of accuracies")
print(
combined.pivot_table(index='hue', columns='ncolors',
values='acc').corr())
p = plot_hue_accuracies(combined)
plot_path = os.path.join(outdir, '%s_image_hue_accuracies.pdf' % scorer)
p.get_figure().savefig(plot_path)
csv_path = os.path.join(outdir, '%s_image_hue_accuracies.csv' % scorer)
combined.to_csv(csv_path, index=False)
# plot as scatter plot of points along the kodak and
# specimen accuracies
specimen_df = observed[['hue', 'acc']].rename(columns={'acc': 'specimen'})
images_df = pd.merge(images_df, specimen_df, on='hue', how='left')
for ncolors in images_df['ncolors'].unique():
for norm in [True, False]:
scatter_plot = plot_hue_accuracies_lm(images_df,
ncolors,
norm=norm)
norm_str = 'normalized' if norm else 'raw'
scatter_plot_path = os.path.join(
outdir, '%s-%d-image-hue-accuracies-scatter-plot-%s.pdf' %
(scorer, ncolors, norm_str))
scatter_plot.savefig(scatter_plot_path)
plt.close('all')
def _get_user_data(self):
db = s_qs.SpecimenQueries(database_path=self.database_path)
df = db.execute_adhoc("""
select users.uniqueid as userid,
specimen_h,
target_h,
specimen_r, specimen_g, specimen_b,
target_r, target_g, target_b,
specimen_lab_l, specimen_lab_a, specimen_lab_b,
target_lab_l, target_lab_a, target_lab_b,
correct
from selectionevents inner join users
on selectionevents.userid = users.id
where users.uniqueId = '%s'
""" % self.userid)
# make sure correct is boolean, sqlite represents as integer, which messes pandas up
# sometimes when using as a predicate for row selection
df['correct'] = df['correct'] == 1
return df
lambda args: cvd_user_accuracy(db, args.input_path, args.output_path),
'average_absolute_filesizes':
lambda args: average_absolute_filesizes(args.input_path, args.
output_path)
}
available_actions_str = ','.join(list(available_actions.keys()))
parser = argparse.ArgumentParser(description=description)
parser.add_argument('database',
type=str,
help='Path to Specimen database file')
parser.add_argument('input_path',
type=str,
help="Path to results relevant for given function")
parser.add_argument(
'output_path',
type=str,
help="Path to save down plot/result relevant for given function")
parser.add_argument('action',
type=str,
help='One of [%s]' % available_actions_str)
args = parser.parse_args()
db = s_qs.SpecimenQueries(database_path=args.database)
if not args.action in available_actions:
raise ValueError("Undefined action: %s, must be one of [%s]" %
(args.action, available_actions_str))
available_actions.get(args.action)(args)
