def compare_params(project_name):
excluded_keys = ['param_name', 'job_name', 'save_path', 'project_path']
if request.method == 'POST':
param_names = request.get_json()['param_names']
print('param_names:')
print(param_names)
else:
param_names = []
param2val_list = []
for param_name in param_names:
param_path = to_param_path(project_name, param_name)
with (param_path / 'param2val.yaml').open('r') as f:
param2val = yaml.load(f, Loader=yaml.FullLoader)
param2val_list.append(param2val)
message = ''
keys = [k for k in param2val_list[0].keys() if k not in excluded_keys]
for key in keys:
param_values = [param2val[key] for param2val in param2val_list]
if len(set(param_values)
) != 1: # param_values differ between configurations
message += '<p><b>{}</b>={}</p>'.format(key, param_values)
return message + '<p>(shown in order of configurations selected)</p>'
def plot(project_name):
"""
is requested in two ways:
1. directly clicking on row in project.html
2. by selecting multiple runs and clicking on "plot" button
in case 1, param_names is retrieved from request object.
in case 2, param_names is retrieved from session object (because of redirect)
check request object first, and only then check session
"""
param_names = request.args.getlist('param_name') or session.get(
'param_names')
# TODO is there a way to plot confidence interval?
# TODO if not, then plot all the individual lines, instead of their average?
# get all patterns (all possible csv file names) - assume each run has same pattern
first_param_path = to_param_path(project_name, param_names[0])
patterns = set([p.name for p in first_param_path.rglob('*.csv')])
if not patterns:
raise LudwigVizNoCsvFound(first_param_path)
else:
print('Detected patterns={}'.format(patterns))
# iterate over unique df file names (e.g. results_a.csv, results_b.csv)
json_charts = []
for pattern in patterns:
print('pattern="{}"'.format(pattern))
# get data frame where each column represents a mean for a particular param_name
try:
data = aggregate_data(project_name, param_names, pattern)
except ValueError: # empty data frames
print(f'WARNING: Failed ot aggregate data for pattern={pattern}')
continue
# make chart
title = pattern.rstrip('.csv').capitalize()
column_name = data.columns[0]
json_chart = make_json_chart(data, column_name, title)
# collect chart
json_charts.append(json_chart)
# get number of reps for each param_name
param_name2n = {
param_name: count_replications(project_name, param_name)
for param_name in param_names
}
return render_template(
'plots.html',
topbar_dict=topbar_dict,
project_name=project_name,
param_names=param_names,
param_name2n=param_name2n,
json_charts=json_charts,
)
def count_replications(project_name, param_name):
return len(list(to_param_path(project_name, param_name).glob('*[!.yaml]')))