def genFeatures(self, t, df, **kwargs):
cols = self.lag_cols + [t]
ld = self.stationId[0].isupper()
self.ld = ld # thread-unsafe
self.target = t # thread-unsafe
if ld is True:
print('London city')
cols.remove('weather_clu')
features = []
for c in cols:
features += [
_ for _ in df.columns if _.find(lag_format(c, '')) != -1
]
features += [
'hour'
] + cols #, 'temperature', 'pressure', 'humidity', 'wind_direction', 'wind_speed']
if ld is False and self.target != 'O3':
features.append('last_rain')
features.remove(t)
if self.target == 'O3':
res1 = []
for f in features:
if f.startswith('weather_clu'):
res1.append('weather' + f[len('weather_clu'):])
else:
res1.append(f)
features = res1
return features
def _gen_from_dict(d):
res = []
for k, v in d.items():
if type(v) == int:
v = range(v + 1)
for i in v:
res.append(lag_format(k, i))
return res
def __init__(self, args, creator=dataset.create_dataset1):
self.args = args
self.dataset = creator(args.data_path, MaxLagging=args.lag)
if args.predict_date is not None:
for gas in ['PM2.5', 'PM10', 'O3']:
for i in range(args.lag + 1):
self.dataset.loc[
self.dataset.utc_time >= args.predict_date,
lag_format(gas, i)] = np.nan
# print('Latest one with non-nan {} in dataset : '.format(lag_format(gas, i)), self.dataset.dropna(subset=[lag_format(gas, i)]).utc_time.max())
print('Latest of dataset: ', self.dataset.utc_time.max())
def genNext(X_test, y_pred, features, ref):
assert len(X_test) == 1
assert ref is not None
##### SLOW ######
# new_X_test = X_test.copy()
# new_X_test.utc_time += pd.DateOffset(hours=1)
# if 'hour' in new_X_test.columns:
# new_X_test['hour'] = (1 + new_X_test['hour']) % 24
# find = False
# if df is not None:
# tmp, find = find_new_X_test(new_X_test)
# if find:
# new_X_test = tmp.copy()
##### SHOULD BE LESS SLOW ######
find = True
new_X_test = ref
new_X_test = new_X_test.reset_index(drop=True)
X_test = X_test.reset_index(drop=True)
for c in features:
match = re.match('(.*?)_lag_(\d+)', c)
if match is not None:
name = match.group(1)
lag_idx = int(match.group(2))
if lag_idx > 1:
col = '{}_lag_{}'.format(name, lag_idx - 1)
if col in X_test.columns:
new_X_test[c] = X_test[col]
elif name in X_test.columns and name != target:
new_X_test[c] = X_test[name]
if not find or not deploy or new_X_test[lag_format(target,
1)].isna().any():
new_X_test[lag_format(target, 1)] = y_pred
# print(np.expm1(new_X_test[ [lag_format(target, i) for i in range(13)] ].iloc[0].values))
return new_X_test
def genFeatures(self, t, df, **kwargs):
Strategy.genFeatures(self, t, df, **kwargs)
cols = self.lag_cols + [t]
if self.stationId[0].isupper():
print('London city')
cols.remove('weather')
features = []
for c in cols:
features += [
_ for _ in df.columns if _.find(lag_format(c, '')) != -1
]
features += [
'hour'
] + cols #, 'temperature', 'pressure', 'humidity', 'wind_direction', 'wind_speed']
features.remove(t)
return features
def genFeatures(self, t, df, **kwargs): #
s = self.stationId
features = list(
self.features[(self.features.stationId == s)
& (self.features.gas == t) &
(self.features.importance > self.thres)].feature)
self_lag = []
for f in features:
match = re.match(r'{}_lag_(\d+)'.format(t), f)
if match is not None:
self_lag.append(int(match.group(1)))
self_lag = set(self_lag)
for i in range(1, max(self_lag)):
if i not in self_lag:
features.append(lag_format(t, i))
print(features)
return features
def genFeatures(self, t, df, **kwargs):
Strategy.genFeatures(self, t, df, **kwargs)
cols = self.lag_cols + [t]
ld = self.ld
# print('??????', self.stationId)
if ld is True:
print('London city')
cols.remove('weather_clu')
features = []
for c in cols:
features += [
_ for _ in df.columns if _.find(lag_format(c, '')) != -1
]
features += [
'hour'
] + cols #, 'temperature', 'pressure', 'humidity', 'wind_direction', 'wind_speed']
if ld is False:
features.append('rain_hours')
features.remove(t)
return features