-
Notifications
You must be signed in to change notification settings - Fork 0
/
nigeria_survey_validation.py
369 lines (281 loc) · 14.4 KB
/
nigeria_survey_validation.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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
"""
specify survey layer (with year) to validate
load s4_surface for same year (use json or fixed args to determine which surface?)
compare point (or buffer zs?) value of surface to survey value for each survey point
output dataframe of survey value, point value, buffer value along with percent differences? (include lat/lon to map errors)
"""
import os
import itertools
import errno
import glob
import time
import datetime
import pandas as pd
import numpy as np
import sklearn.metrics
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from utils.settings_builder import Settings
def make_dir(path):
try:
os.makedirs(path)
except OSError as exception:
if exception.errno != errno.EEXIST:
raise
class ConfusionMatrix():
def __init__(self, true, pred, proba=False, thresh=0.5):
self.true = true
self.raw_pred = pred
if proba:
self.pred = (pred > thresh).astype(int)
self.tp = sum((true == 1) & (self.pred == 1))
self.fn = sum((true == 1) & (self.pred == 0))
self.tn = sum((true == 0) & (self.pred == 0))
self.fp = sum((true == 0) & (self.pred == 1))
self.cm = (self.tp, self.fn, self.tn, self.fp)
self.gen_rates()
self.gen_performance_measures()
def run(self):
tpr, fnr, tnr, fpr = self.gen_rates()
accuracy, precision, recall, f1 = self.gen_performance_measures()
out = {
"tp": tpr, "fn": fnr, "tn": tnr, "fp": fpr,
"accuracy": accuracy, "precision": precision, "recall": recall, "f1": f1
}
return out
def gen_rates(self):
tpr = self.calc_tp_rate()
fnr = self.calc_fn_rate()
tnr = self.calc_tn_rate()
fpr = self.calc_fp_rate()
return (tpr, fnr, tnr, fpr)
def gen_performance_measures(self):
accuracy = sklearn.metrics.accuracy_score(self.true, self.pred)
precision = sklearn.metrics.precision_score(self.true, self.pred)
recall = sklearn.metrics.recall_score(self.true, self.pred)
f1 = sklearn.metrics.f1_score(self.true, self.pred)
return (accuracy, precision, recall, f1)
def calc_tp_rate(self):
try:
return self.tp / float(self.tp+self.fn)
except:
return None
def calc_fn_rate(self):
try:
return self.fn / float(self.fn+self.tp)
except:
return None
def calc_tn_rate(self):
try:
return self.tn / float(self.tn+self.fp)
except:
return None
def calc_fp_rate(self):
try:
return self.fp / float(self.fp+self.tn)
except:
return None
timestamp = datetime.datetime.fromtimestamp(int(time.time())).strftime(
'%Y_%m_%d_%H_%M_%S')
json_path_list = glob.glob("../*acled/settings/nigeria_acled.json")
# group versions of same temporal
version_groups = {}
for json_path in json_path_list:
temporal = json_path.split("/")[1].split("_")[1]
s = Settings()
s.load(json_path)
tag = "{}_{}".format(s.config["version"], s.config["predict_tag"])
if temporal not in version_groups:
version_groups[temporal] = []
version_groups[temporal].append(json_path)
cm_list = []
thresh_val_list = [0.3, 0.35, 0.4, 0.45, 0.5]
for ix, temporal in enumerate(version_groups.keys()):
# ==============
# WARNING: THIS WILL NOT CURRENTLY WORK IF YOU HAVE MORE THAN ONE PARAM COMBO
# it will just put all the params from same temporal group in same plot
# ==============
# init roc curve plot
plt.figure(1)
# init prc plot
plt.figure(2)
auc_scores = []
for iy, json_path in enumerate(version_groups[temporal]):
# for json_path in json_path_list:
batch_id = "_".join(json_path.split("/")[1].split("_")[:2])
print(batch_id)
s = Settings()
s.load(json_path)
validation_dir = os.path.join(s.base_path, "output/s4_survey_validation")
make_dir(validation_dir)
predict_settings = s.data["predict"]["acled"]
predict_hash = s.build_hash(predict_settings, nchar=7)
version_tag = s.config["version"]
predict_tag = s.config["predict_tag"]
tasks = s.hashed_iter()
for ix, (param_hash, _params) in enumerate(tasks):
combo_temporal = "{}_{}".format(predict_settings["imagery"][0], predict_settings["sample"])
print("Running: {} : {} - {} {}".format(param_hash, combo_temporal, version_tag, predict_tag))
survey_prediction_path = os.path.join(s.base_path, "output/s1_predict/predict_{}_{}_{}_{}.csv".format(param_hash, predict_hash, version_tag, predict_tag))
survey_prediction_df = pd.read_csv(survey_prediction_path)
survey_prediction_df = survey_prediction_df.loc[survey_prediction_df["group"] == "original"].copy(deep=True)
survey_prediction_df = survey_prediction_df[[i for i in survey_prediction_df.columns if not i.startswith("feat_")]]
survey_prediction_df['index'] = survey_prediction_df['data_id']
survey_prediction_df.set_index('index', inplace=True)
sample_path = os.path.join(s.base_path, "data/grid/sample_trim_{}_{}.csv".format(s.build_hash(s.static), version_tag))
sample_df = pd.read_csv(sample_path)
sample_df = sample_df.loc[sample_df["group"] == "original"].copy(deep=True)
sample_df = sample_df[["data_id", "type", "drop"]]
sample_df['index'] = sample_df['data_id']
sample_df.set_index('index', inplace=True)
survey_df = sample_df.join(survey_prediction_df, how="left", lsuffix="_sample", rsuffix="_predict")
survey_df["binary"] = (survey_df["fatalities"] > 0).astype(int)
if "lon" not in survey_df.columns:
survey_df["lon"] = survey_df["longitude"]
if "lat" not in survey_df.columns:
survey_df["lat"] = survey_df["latitude"]
print("Total Survey points: {}".format(len(survey_df)))
survey_df = survey_df.loc[survey_df["type"] == "val"]
print("Non-duplicate validation points: {}".format(len(survey_df)))
y_true = survey_df["binary"]
y_prob = survey_df["proba_1_1"]
# ----------------------------------------------
# generate confusion matrix at varying thresholds
tmp_survey_df = survey_df.copy(deep=True)
survey_df_list = []
for thresh_val in thresh_val_list:
validation_fname = "val_{}_{}_{}_{}".format(param_hash, predict_hash, version_tag, predict_tag)
# validation_base = os.path.join(validation_dir, validation_fname)
stats = ConfusionMatrix(y_true, y_prob, proba=True, thresh=thresh_val)
tmp_cm_data = stats.run()
cm_data = {}
for k in tmp_cm_data.keys():
cm_data[k] = round(tmp_cm_data[k], 3)
cm_data["id"] = validation_fname
cm_data["temporal"] = combo_temporal
cm_data["version"] = version_tag
cm_data["thresh"] = thresh_val
cm_list.append(cm_data)
y_pred = stats.pred
tmp_survey_df["thresh_{}_match".format(thresh_val)] = (y_true == y_pred).astype(int)
col_name = "thresh_{}_confusion".format(thresh_val)
tmp_survey_df[col_name] = None
tmp_survey_df.loc[((y_true == 1) & (y_pred == 1)), col_name] = "tp"
tmp_survey_df.loc[((y_true == 0) & (y_pred == 0)), col_name] = "tn"
tmp_survey_df.loc[((y_true == 0) & (y_pred == 1)), col_name] = "fp"
tmp_survey_df.loc[((y_true == 1) & (y_pred == 0)), col_name] = "fn"
survey_df_list.append(tmp_survey_df)
final_survey_df = pd.concat(survey_df_list)
final_survey_path = os.path.join(validation_dir, "{}_{}_{}_{}_{}_thresh-{}_survey.csv".format(timestamp, param_hash, predict_hash, version_tag, predict_tag, thresh_val))
final_survey_df.to_csv(final_survey_path, index=False)
# ----------------------------------------------
# generate roc
auc = sklearn.metrics.roc_auc_score(y_true, y_prob)
auc_scores.append(auc)
fpr, tpr, _ = sklearn.metrics.roc_curve(y_true, y_prob)
# 1:1 line (noskill) data
ns_probs = [0 for _ in range(len(y_true))]
ns_auc = sklearn.metrics.roc_auc_score(y_true, ns_probs)
ns_fpr, ns_tpr, _ = sklearn.metrics.roc_curve(y_true, ns_probs)
# plt.figure()
plt.figure(1)
# plt.plot(ns_fpr, ns_tpr, linestyle='--', label='No Skill')
plt.plot(fpr, tpr, marker='.', label='Batch {} (AUC {})'.format(iy, round(auc,3)))
# plt.xlabel('False Positive Rate')
# plt.ylabel('True Positive Rate')
# plt.legend(loc='lower right')
# plt.title("ROC Curve")
# plot_path = os.path.join(validation_dir, "{}_{}_{}_{}_{}_roc.png".format(timestamp, param_hash, predict_hash, version_tag, predict_tag))
# plt.savefig(plot_path)
print('No Skill: ROC AUC=%.3f' % (ns_auc))
print('Actual: ROC AUC=%.3f' % (auc))
# ----------------------------------------------
# generate prc
prc_precision, prc_recall, thresholds = sklearn.metrics.precision_recall_curve(y_true, y_prob)
# no_skill = len(y_true[y_true==1]) / float(len(y_true))
# plt.figure()
plt.figure(2)
plt.plot(prc_recall, prc_precision, marker='.', label='Batch {}'.format(iy))
# plt.ylim(0.45, 1.05)
# plt.xlabel('Recall')
# plt.ylabel('Precision')
# plt.legend()
# plt.title("PRC Curve")
# plot_path = os.path.join(validation_dir, "{}_{}_{}_{}_{}_prc.png".format(timestamp, param_hash, predict_hash, version_tag, predict_tag))
# plt.savefig(plot_path)
if len(temporal) == 2:
title_temporal = "201{} imagery to predict 201{} conflict".format(*temporal)
elif len(temporal) == 4:
title_temporal = "201{} h{} imagery to predict 201{} h{} conflict".format(*temporal)
else:
raise Exception("Invalid temporal string ({})".format(temporal))
plt.figure(1)
plt.plot(ns_fpr, ns_tpr, linestyle='--', label='No Skill (AUC 0.5)')
plt.title("ROC Curve - Mean AUC: {}".format(round(np.mean(auc_scores), 3)))
plt.suptitle("{}".format(title_temporal))
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.legend(loc='lower right')
plot_path = os.path.join(validation_dir, "{}_{}_{}_{}_roc.png".format(timestamp, param_hash, predict_hash, temporal))
plt.savefig(plot_path)
plt.clf()
plt.figure(2)
plt.plot([0, 1], [0.5, 0.5], linestyle='--', label='No Skill')
plt.title("Precision-Recall Curve")
plt.suptitle("{}".format(title_temporal))
plt.ylim(0.45, 1.05)
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.legend()
plot_path = os.path.join(validation_dir, "{}_{}_{}_{}_prc.png".format(timestamp, param_hash, predict_hash, temporal))
plt.savefig(plot_path)
plt.clf()
final_summary_df = pd.DataFrame(cm_list)
final_summary_df = final_summary_df[["id", "temporal", "version", "thresh", "tp", "tn", "fp", "fn", "accuracy", "precision", "recall", "f1"]]
final_summary_path = os.path.join(validation_dir, "{}_summary.csv".format(timestamp))
final_summary_df.to_csv(final_summary_path, index=False)
# full confusion metrics only
full_cm_summary_df = final_summary_df[["temporal", "version", "thresh", "tp", "tn", "fp", "fn"]].copy(deep=True)
full_cm_summary_path = os.path.join(validation_dir, "{}_summary_full_cm_only.csv".format(timestamp))
full_cm_summary_df.to_csv(full_cm_summary_path, index=False)
# full performance metrics only
full_pm_summary_df = final_summary_df[["temporal", "version", "thresh", "accuracy", "precision", "recall", "f1"]].copy(deep=True)
full_pm_summary_path = os.path.join(validation_dir, "{}_summary_full_pm_only.csv".format(timestamp))
full_pm_summary_df.to_csv(full_pm_summary_path, index=False)
# temporal grouped confusion metrics only
temporal_cm_summary_df = final_summary_df[["temporal", "thresh", "tp", "tn", "fp", "fn"]].copy(deep=True)
temporal_cm_summary_df = temporal_cm_summary_df.groupby(["temporal", "thresh"]).agg({"tp": "mean", "tn": "mean", "fp": "mean", "fn": "mean"}).reset_index()
temporal_cm_summary_path = os.path.join(validation_dir, "{}_summary_temporal_cm_only.csv".format(timestamp))
temporal_cm_summary_df.to_csv(temporal_cm_summary_path, index=False)
# temporal grouped performance metrics only
temporal_pm_summary_df = final_summary_df[["temporal", "thresh", "accuracy", "precision", "recall", "f1"]].copy(deep=True)
temporal_pm_summary_df = temporal_pm_summary_df.groupby(["temporal", "thresh"]).agg({"accuracy": "mean", "precision": "mean", "recall": "mean", "f1": "mean"}).reset_index()
temporal_pm_summary_path = os.path.join(validation_dir, "{}_summary_temporal_pm_only.csv".format(timestamp))
temporal_pm_summary_df.to_csv(temporal_pm_summary_path, index=False)
# ===================================================================
# raise Exception("End of automated code")
# from nigeria_survey_validation import *
# final_summary_df.groupby("thresh").agg({"tp":"mean", "fn":"mean", "fp":"mean", "tn": "mean"})
# final_summary_df.groupby("thresh").agg({"accuracy":"mean", "precision":"mean", "recall":"mean", "f1": "mean"})
# w = final_summary_df.copy(deep=True)
# w["version"] = w.apply(lambda x: x["id"].split("_")[-2] , axis=1)
# w.groupby("version").agg({"accuracy":"mean", "precision":"mean", "recall":"mean"})
# y = w.loc[w["thresh"] == 0.35].copy(deep=True)
# z = y[["id", "temporal", "accuracy", "precision", "recall"]].copy(deep=True)
# z.groupby("temporal").agg({"accuracy":"mean", "precision":"mean", "recall":"mean"})
# ===================================================================
"""
# check static hashes
from __future__ import print_function, division
from utils.settings_builder import Settings
from utils.data_prep import *
import glob
hashes = {}
for json_path in glob.glob("/home/userv/Desktop/laboi_batches_best/*.json"):
s = Settings()
s.load(json_path)
static_hash = s.build_hash(s.static)
hashes[static_hash] = {"imagery": s.static["sample_definition"]["acled"]["imagery"], "sample": s.static["sample_definition"]["acled"]["sample"]}
print(static_hash, hashes[static_hash])
"""