def predict_demand(regr, X, y):
"""
Predict tM3 according to the model t^M3 = tM1 + delta
and aggregate the predicted tM3 to obtain the predicted demand
regr is a regression model that can predict delays delta = tM3 - tM1
X is a matrix of features with X[:, 0] being tM1
y is a vector with tM3
"""
params = regr.get_params()
interval = params['interval']
tz = params['tz']
tm3 = pd.to_datetime(y)
# transform y in data format
y_true = atddm.binarrivals(tm3, interval=interval, tz=tz).fillna(0)
# bin observed tM3 to obtain observed demand
y_pred = atddm.binarrivals(regr.predict(X), interval=interval,
tz=tz).fillna(0)
# bin predicted tM3 to obtain predicted demand
combined_indx = y_true.index.union(y_pred.index)
y_true = y_true.reindex(index=combined_indx).fillna(0)
y_pred = y_pred.reindex(index=combined_indx).fillna(0)
# reindex predicted demand as observed one to avoid length mismatch
return y_true, y_pred
ENDDT = pd.Timestamp(ENDDT)
# ENDDT = BEGDT + pd.Timedelta(21, 'D')
INTERVAL = 10
dd = atddm.load(subset=CODES)
m3_bin = {}
nairp = len(CODES)
CODES.sort()
for code in CODES:
indx = pd.date_range(start=BEGDT,
end=ENDDT,
freq=str(INTERVAL) + 'min',
tz=TZONES[code])
m3_bin[code] = atddm.binarrivals(dd[code].M3_FL240,
interval=INTERVAL,
tz=TZONES[code])[indx].fillna(0)
if TRGT == 'talk':
f, axes = plt.subplots(2, nairp // 2, sharey=False)
else:
f, axes = plt.subplots(nairp // 2, 2, sharey=False)
for ax, code in zip(axes.flatten(), CODES):
ts = m3_bin[code].fillna(0)
ts = ts.diff().dropna()
freq = 60*ts.index.freq.delta.components.hours +\
ts.index.freq.delta.components.minutes
day0 = ts.index[0]
days = int(len(ts) / (24 * 60 / freq))
begdays = [day0 + pd.Timedelta(i, 'D') for i in range(0, days)]
for t in range(0, 24 * 60, INTERVAL)
]
end_times = [
midnight + pd.Timedelta((INTERVAL + t) * 60 - 1, unit='s')
for t in range(0, 24 * 60, INTERVAL)
]
slices = [(a.time(), b.time()) for a, b in zip(sta_times, end_times)]
for code in CODES:
df = dd[code]
tz = TZONES[code]
indx = pd.date_range(start=BEGDT,
end=ENDDT,
freq=str(INTERVAL) + 'min',
tz=tz)
m3_bin[code] = atddm.binarrivals(df.M3_FL240, interval=INTERVAL,
tz=tz)[indx].fillna(0)
m3_bin[code].index = m3_bin[code].index.tz_localize(None)
m1_bin[code] = atddm.binarrivals(df.M1_FL240, interval=INTERVAL,
tz=tz)[indx].fillna(0)
m1_bin[code].index = m1_bin[code].index.tz_localize(None)
tmp = df.M1_FL240 + np.array(df.delay.sample(n=len(df), replace=True))
psra_bin[code] = atddm.binarrivals(tmp, interval=INTERVAL,
tz=tz)[indx].fillna(0)
psra_bin[code].index = psra_bin[code].index.tz_localize(None)
daily = {}
daily_psra = {}
for code in CODES:
tz = TZONES[code]
daily[code] = atddm.daily_avg(m3_bin[code], tz=tz)
daily[code].index = daily[code].index.tz_localize(None)