-
Notifications
You must be signed in to change notification settings - Fork 28
/
evaluate_performance.py
160 lines (141 loc) · 9.18 KB
/
evaluate_performance.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
import os
import json
import numpy as np
import pandas as pd
from collections import defaultdict
from dataclasses import fields
import mlflow
import hydra
from hydra.utils import to_absolute_path
from omegaconf import OmegaConf, DictConfig
import utils.evaluation as eval_tools
@hydra.main(config_path='configs/eval', config_name='run3')
def main(cfg: DictConfig) -> None:
mlflow.set_tracking_uri(f"file://{to_absolute_path(cfg.path_to_mlflow)}")
# setting paths
# path_to_weights_taus = to_absolute_path(cfg.path_to_weights_taus) if cfg.path_to_weights_taus is not None else None
# path_to_weights_vs_type = to_absolute_path(cfg.path_to_weights_vs_type) if cfg.path_to_weights_vs_type is not None else None
path_to_artifacts = to_absolute_path(f'{cfg.path_to_mlflow}/{cfg.experiment_id}/{cfg.run_id}/artifacts/')
output_json_path = f'{path_to_artifacts}/performance.json'
# init Discriminator() class from filtered input configuration
field_names = set(f_.name for f_ in fields(eval_tools.Discriminator))
init_params = {k:v for k,v in cfg.discriminator.items() if k in field_names}
if 'wp_thresholds' in init_params:
if not (isinstance(init_params['wp_thresholds'], DictConfig) or isinstance(init_params['wp_thresholds'], dict)):
if isinstance(init_params['wp_thresholds'], str): # assume that it's the filename to read WPs from
with open(f"{path_to_artifacts}/{init_params['wp_thresholds']}", 'r') as f:
wp_thresholds = json.load(f)
init_params['wp_thresholds'] = wp_thresholds[cfg['vs_type']] # pass laoded dict with thresholds to Discriminator() class
else:
raise RuntimeError(f"Expect `wp_thresholds` argument to be either dict-like or str, but got the type: {type(init_params['wp_thresholds'])}")
else:
wp_thresholds = None
discriminator = eval_tools.Discriminator(**init_params)
# construct branches to be read from input files
input_branches = OmegaConf.to_object(cfg.input_branches)
if ((_b:=discriminator.pred_column) is not None) and (cfg.path_to_pred is None):
input_branches.append(_b)
if (discriminator.wp_column) is not None:
if 'wp_name_to_index_map' in cfg['discriminator']: # append all branches for multiclass WP models
for tau_type in cfg['discriminator']['wp_name_to_index_map'].keys():
input_branches.append(cfg['discriminator']['wp_column_prefix'] + tau_type)
else: # append only wp_column branch for binary WP model
input_branches.append(discriminator.wp_column)
# loop over input samples
df_list = []
print()
for sample_alias, tau_types in cfg.input_samples.items():
input_files, pred_files, target_files = eval_tools.prepare_filelists(sample_alias, cfg.path_to_input, cfg.path_to_pred, cfg.path_to_target, path_to_artifacts)
# loop over all input files per sample with associated predictions/targets (if present) and combine together into df
print(f'[INFO] Creating dataframe for sample: {sample_alias}')
for input_file, pred_file, target_file in zip(input_files, pred_files, target_files):
df = eval_tools.create_df(input_file, input_branches, pred_file, target_file, None, # weights functionality is WIP
cfg.discriminator.pred_column_prefix, cfg.discriminator.target_column_prefix)
gen_selection = ' or '.join([f'(gen_{tau_type}==1)' for tau_type in tau_types]) # gen_* are constructed in `add_targets()`
df = df.query(gen_selection)
df_list.append(df)
df_all = pd.concat(df_list)
# apply selection
if cfg['cuts'] is not None:
df_all = df_all.query(cfg.cuts)
if cfg['WPs_to_require'] is not None:
for wp_vs_type, wp_name in cfg['WPs_to_require'].items():
if cfg['discriminator']['wp_from']=='wp_column':
wp = cfg['discriminator']['wp_name_to_index_map'][wp_vs_type][wp_name]
wp_column = f"{cfg['discriminator']['wp_column_prefix']}{wp_vs_type}"
flag = 1 << wp
df_all = df_all[np.bitwise_and(df_all[wp_column], flag) != 0]
else:
if wp_thresholds is not None: # take thresholds from previously loaded json
wp_thr = wp_thresholds[wp_vs_type][wp_name]
elif cfg['discriminator']['wp_thresholds_map'] is not None: # take thresholds from discriminator cfg
wp_thr = cfg['discriminator']['wp_thresholds_map'][wp_vs_type][wp_name]
else:
raise RuntimeError('WP thresholds either from wp_column, or wp_thresholds_map, or via input json file are not provided.')
wp_cut = f"{cfg['discriminator']['pred_column_prefix']}{wp_vs_type} > {wp_thr}"
df_all = df_all.query(wp_cut)
# # inverse scaling
# df_all['tau_pt'] = df_all.tau_pt*(1000 - 20) + 20
# dump curves' data into json file
json_exists = os.path.exists(output_json_path)
json_open_mode = 'r+' if json_exists else 'w'
with open(output_json_path, json_open_mode) as json_file:
if json_exists: # read performance data to append additional info
performance_data = json.load(json_file)
else: # create dictionary to fill with data
performance_data = {'name': discriminator.name, 'metrics': defaultdict(list)}
# loop over pt bins
print(f'\n{discriminator.name}')
for dm_bin in cfg.dm_bins:
for eta_index, (eta_min, eta_max) in enumerate(cfg.eta_bins):
for pt_index, (pt_min, pt_max) in enumerate(cfg.pt_bins):
# apply pt/eta/dm bin selection
df_cut = df_all.query(f'tau_pt >= {pt_min} and tau_pt < {pt_max} and abs(tau_eta) >= {eta_min} and abs(tau_eta) < {eta_max} and tau_decayMode in {dm_bin}')
if df_cut.shape[0] == 0:
print("Warning: bin with pt ({}, {}) and eta ({}, {}) and DMs {} is empty.".format(pt_min, pt_max, eta_min, eta_max, dm_bin))
continue
print(f'\n-----> pt bin: [{pt_min}, {pt_max}], eta bin: [{eta_min}, {eta_max}], DM bin: {dm_bin}')
print('[INFO] counts:\n', df_cut[['gen_tau', f'gen_{cfg.vs_type}']].value_counts())
# create roc curve and working points
roc, wp_roc = discriminator.create_roc_curve(df_cut)
if roc is not None:
# prune the curve
roc = roc.prune(tpr_decimals=cfg['roc_prune_decimal'][cfg['vs_type']])
if roc.auc_score is not None:
print(f'[INFO] ROC curve done, AUC = {roc.auc_score:.6f}')
# loop over [ROC curve, ROC curve WP] for a given discriminator and store its info into dict
for curve_type, curve in zip(['roc_curve', 'roc_wp'], [roc, wp_roc]):
if curve is None: continue
if json_exists and curve_type in performance_data['metrics'] \
and (existing_curve := eval_tools.select_curve(performance_data['metrics'][curve_type],
pt_min=pt_min, pt_max=pt_max, eta_min=eta_min, eta_max=eta_max, dm_bin=dm_bin, vs_type=cfg.vs_type,
dataset_alias=cfg.dataset_alias)) is not None:
print(f'[INFO] Found already existing curve (type: {curve_type}) in json file for a specified set of parameters: will overwrite it.')
performance_data['metrics'][curve_type].remove(existing_curve)
curve_data = {
'pt_min': pt_min, 'pt_max': pt_max,
'eta_min': eta_min, 'eta_max': eta_max,
'dm_bin': list(dm_bin),
'vs_type': cfg.vs_type,
'dataset_alias': cfg.dataset_alias,
'auc_score': curve.auc_score,
'false_positive_rate': eval_tools.FloatList(curve.pr[0, :].tolist()),
'true_positive_rate': eval_tools.FloatList(curve.pr[1, :].tolist()),
}
if curve.thresholds is not None:
curve_data['thresholds'] = eval_tools.FloatList(curve.thresholds.tolist())
if curve.pr_err is not None:
curve_data['false_positive_rate_up'] = eval_tools.FloatList(curve.pr_err[0, 0, :].tolist())
curve_data['false_positive_rate_down'] = eval_tools.FloatList(curve.pr_err[0, 1, :].tolist())
curve_data['true_positive_rate_up'] = eval_tools.FloatList(curve.pr_err[1, 0, :].tolist())
curve_data['true_positive_rate_down'] = eval_tools.FloatList(curve.pr_err[1, 1, :].tolist())
# append data for a given curve_type and pt bin
if curve_type not in performance_data['metrics']:
performance_data['metrics'][curve_type] = []
performance_data['metrics'][curve_type].append(curve_data)
json_file.seek(0)
json_file.write(json.dumps(performance_data, indent=4, cls=eval_tools.CustomJsonEncoder))
json_file.truncate()
print()
if __name__ == '__main__':
main()