def test_ar_forecast(self):
""" using AR to extend X to be longer than Y so must train earlier """
cfg = TSConfig()
cfg.register_dataset(TARGET_DF, 'y', 'target')
cfg.register_dataset(TRAIN_DF, 'x', 'predictor')
cfg.add_AR([2], dataset='x')
cfg.stack()
arex = Arex(model=mm, data_config=cfg)
arex.forecast(t_plus=2, pred_start="2019-02-26", pred_end="2019-03-05",
training="roll", window=2, t_known=False)
log = arex.log
assert to_ymd(log[0]['time']) == '2019-02-26'
assert to_ymd(log[1]['time']) == '2019-03-05'
info = log[0]
assert to_ymd(info['X_train'][0]) == '2019-01-29'
assert to_ymd(info['X_train'][1]) == '2019-02-05'
assert to_ymd(info['y_train'][0]) == '2019-02-12'
assert to_ymd(info['y_train'][1]) == '2019-02-19'
assert to_ymd(info['X_pred']) == '2019-02-26'
assert info['sizes'][0] == (2, 2)
info = log[1]
assert to_ymd(info['X_train'][0]) == '2019-01-29'
assert to_ymd(info['X_train'][1]) == '2019-02-05'
assert to_ymd(info['y_train'][0]) == '2019-02-12'
assert to_ymd(info['y_train'][1]) == '2019-02-19'
assert to_ymd(info['X_pred']) == '2019-03-05'
assert info['sizes'][0] == (2, 2)
def test_t_known_expand_forecast(self):
cfg = TSConfig()
cfg.register_dataset(TARGET_DF, 'y', 'target')
cfg.register_dataset(TRAIN_DF, 'x', 'predictor')
cfg.stack()
arex = Arex(model=mm, data_config=cfg)
arex.forecast(t_plus=2, pred_start="2019-02-12", pred_end="2019-02-19",
training="expand", window=2, t_known=True)
log = arex.log
assert to_ymd(log[0]['time']) == '2019-02-12'
assert to_ymd(log[1]['time']) == '2019-02-19'
info = log[0]
assert to_ymd(info['X_train'][0]) == '2019-01-22'
assert to_ymd(info['X_train'][1]) == '2019-01-29'
assert to_ymd(info['y_train'][0]) == '2019-02-05'
assert to_ymd(info['y_train'][1]) == '2019-02-12'
assert to_ymd(info['X_pred']) == '2019-02-12'
assert info['sizes'][0] == (2, 1)
info = log[1]
assert to_ymd(info['X_train'][0]) == '2019-01-22'
assert to_ymd(info['X_train'][1]) == '2019-02-05'
assert to_ymd(info['y_train'][0]) == '2019-02-05'
assert to_ymd(info['y_train'][1]) == '2019-02-19'
assert to_ymd(info['X_pred']) == '2019-02-19'
assert info['sizes'][0] == (3, 1)
def test_simple_nowcast(self):
cfg = TSConfig()
cfg.register_dataset(TARGET_DF, 'y', 'target')
cfg.register_dataset(TRAIN_DF, 'x', 'predictor')
cfg.stack()
arex = Arex(model=mm, data_config=cfg)
arex.nowcast(pred_start="2019-02-12", pred_end="2019-02-19",
training="roll", window=2)
log = arex.log
assert to_ymd(log[0]['time']) == '2019-02-12'
assert to_ymd(log[1]['time']) == '2019-02-19'
info = log[0]
assert to_ymd(info['X_train'][0]) == '2019-01-29'
assert to_ymd(info['X_train'][1]) == '2019-02-05'
assert to_ymd(info['y_train'][0]) == '2019-01-29'
assert to_ymd(info['y_train'][1]) == '2019-02-05'
assert to_ymd(info['X_pred']) == '2019-02-12'
assert info['sizes'][0] == (2, 1)
def test_window_too_big(self):
cfg = TSConfig()
cfg.register_dataset(TARGET_DF, 'y', 'target')
cfg.register_dataset(TRAIN_DF, 'x', 'predictor')
cfg.add_AR([2], dataset='x')
cfg.stack()
arex = Arex(model=mm, data_config=cfg)
with pytest.raises(ValueError):
arex.nowcast(pred_start="2019-02-19", pred_end="2019-03-05",
training="expand", window=8)
arex.nowcast(pred_start="2019-02-19", pred_end="2019-03-05",
training="expand", window=7)
log = arex.log
assert to_ymd(log[0]['X_train'][0]) == '2019-01-01'
from nowcast.datasets import CDCLoader, gt_loader
from nowcast.models import Argo
from nowcast.utils.metrics import RMSE, Corr
# load data
cdcl = CDCLoader('./ILI_MA.csv', ili_version='unweighted')
ili = cdcl.load_state('MA')
gt = gt_loader('./GTdata_MA.csv')
# configure data with 52 ILI lag terms
dc = TSConfig()
dc.register_dataset(ili, name='ILI', type='target')
dc.register_dataset(gt, name='Trends', type='predictor')
dc.add_AR(range(1, 53), dataset='ILI', var_names=['%ILI'])
dc.stack()
# use the standard ARGO model
base_mod = Argo()
arex = Arex(model=base_mod, data_config=dc)
out = arex.nowcast(pred_start='2017-01-01',
pred_end='2018-04-22',
training='roll',
window=104,
pred_name='ARGO')
predictions = pd.merge(ili, out)
print(predictions)
print('RMSE:', RMSE(predictions['%ILI'], predictions['ARGO']))
print('Correlation:', Corr(predictions['%ILI'], predictions['ARGO']))