def test_prediction_percentiles(iclaims_training_data, prediction_percentiles):
df = iclaims_training_data
lgt = LGTFull(
response_col='claims',
date_col='week',
seasonality=52,
seed=8888,
prediction_percentiles=prediction_percentiles,
)
if not prediction_percentiles:
p_labels = ['_5', '', '_95']
else:
p_labels = ['_5', '_10', '', '_95']
lgt.fit(df)
predicted_df = lgt.predict(df)
expected_columns = ['week'] + ["prediction" + p for p in p_labels]
assert predicted_df.columns.tolist() == expected_columns
assert predicted_df.shape[0] == df.shape[0]
predicted_df = lgt.predict(df, decompose=True)
predicted_components = [
'prediction', PredictedComponents.TREND.value,
PredictedComponents.SEASONALITY.value,
PredictedComponents.REGRESSION.value
]
expected_columns = ['week']
for pc in predicted_components:
for p in p_labels:
expected_columns.append(pc + p)
assert predicted_df.columns.tolist() == expected_columns
assert predicted_df.shape[0] == df.shape[0]
def test_lgt_full_reproducibility(synthetic_data, estimator_type, regressor_signs, seasonality):
train_df, test_df, coef = synthetic_data
lgt_first = LGTFull(
response_col='response',
date_col='week',
regressor_col=train_df.columns.tolist()[2:],
regressor_sign=regressor_signs,
prediction_percentiles=[5, 95],
seasonality=seasonality,
num_warmup=50,
verbose=False,
estimator_type=estimator_type
)
# first fit and predict
lgt_first.fit(train_df)
posteriors_first = copy(lgt_first._posterior_samples)
predict_df_first = lgt_first.predict(test_df)
regression_out_first = lgt_first.get_regression_coefs()
# second fit and predict
# note a new instance must be created to reset the seed
# note both fit and predict contain random generation processes
lgt_second = LGTFull(
response_col='response',
date_col='week',
regressor_col=train_df.columns.tolist()[2:],
regressor_sign=regressor_signs,
prediction_percentiles=[5, 95],
seasonality=seasonality,
num_warmup=50,
verbose=False,
estimator_type=estimator_type
)
lgt_second.fit(train_df)
posteriors_second = copy(lgt_second._posterior_samples)
predict_df_second = lgt_second.predict(test_df)
regression_out_second = lgt_second.get_regression_coefs()
# assert same posterior keys
assert set(posteriors_first.keys()) == set(posteriors_second.keys())
# assert posterior draws are reproducible
for k, v in posteriors_first.items():
assert np.allclose(posteriors_first[k], posteriors_second[k])
# assert identical regression columns
# this is also checked in posterior samples, but an extra layer just in case
# since this one very commonly retrieved by end users
assert regression_out_first.equals(regression_out_second)
# assert prediction is reproducible
assert predict_df_first.equals(predict_df_second)
def test_lgt_full_fit(synthetic_data, seasonality, estimator_type):
train_df, test_df, coef = synthetic_data
args = {
'response_col': 'response',
'date_col': 'week',
'prediction_percentiles': [5, 95],
'seasonality': seasonality,
'verbose': False,
'estimator_type': estimator_type
}
if issubclass(estimator_type, StanEstimator):
expected_num_parameters = 11
args.update({'num_warmup': 50})
else:
# no `lp__` in pyro
expected_num_parameters = 10
args.update({'num_steps': 10})
if seasonality == 52:
expected_num_parameters += 2
lgt = LGTFull(**args)
lgt.fit(train_df)
predict_df = lgt.predict(test_df)
expected_columns = ['week', 'prediction_5', 'prediction', 'prediction_95']
expected_shape = (51, len(expected_columns))
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert len(lgt._posterior_samples) == expected_num_parameters
def test_lgt_non_seasonal_fit_pyro(synthetic_data):
train_df, test_df, coef = synthetic_data
lgt = LGTFull(response_col='response',
date_col='week',
estimator_type=PyroEstimatorVI,
num_steps=10)
lgt.fit(train_df)
predict_df = lgt.predict(test_df)
expected_columns = ['week', 'prediction']
expected_shape = (51, len(expected_columns))
expected_num_parameters = 10 # no `lp__` in pyro
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert len(lgt._posterior_samples) == expected_num_parameters
def test_lgt_full_with_regression(synthetic_data, estimator_type, regressor_signs):
train_df, test_df, coef = synthetic_data
if issubclass(estimator_type, StanEstimator):
lgt = LGTFull(
response_col='response',
date_col='week',
regressor_col=train_df.columns.tolist()[2:],
regressor_sign=regressor_signs,
prediction_percentiles=[5, 95],
seasonality=52,
num_warmup=50,
verbose=False,
estimator_type=estimator_type
)
elif issubclass(estimator_type, PyroEstimator):
lgt = LGTFull(
response_col='response',
date_col='week',
regressor_col=train_df.columns.tolist()[2:],
regressor_sign=regressor_signs,
prediction_percentiles=[5, 95],
seasonality=52,
num_steps=10,
verbose=False,
estimator_type=estimator_type
)
else:
return None
lgt.fit(train_df)
predict_df = lgt.predict(test_df)
regression_out = lgt.get_regression_coefs()
num_regressors = regression_out.shape[0]
expected_columns = ['week', 'prediction_5', 'prediction', 'prediction_95']
expected_shape = (51, len(expected_columns))
expected_regression_shape = (6, 3)
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert regression_out.shape == expected_regression_shape
assert num_regressors == len(train_df.columns.tolist()[2:])
def test_lgt_non_seasonal_fit(synthetic_data, estimator_type):
train_df, test_df, coef = synthetic_data
lgt = LGTFull(
response_col='response',
date_col='week',
estimator_type=estimator_type,
num_warmup=50,
)
lgt.fit(train_df)
predict_df = lgt.predict(test_df)
expected_columns = ['week', 'prediction']
expected_shape = (51, len(expected_columns))
expected_num_parameters = 11
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert len(lgt._posterior_samples) == expected_num_parameters
def test_lgt_full_fit(synthetic_data, seasonality, estimator_type):
train_df, test_df, coef = synthetic_data
args = {
'response_col': 'response',
'date_col': 'week',
'prediction_percentiles': [5, 95],
'seasonality': seasonality,
'verbose': False,
'estimator_type': estimator_type,
}
if issubclass(estimator_type, StanEstimator):
args.update({'num_warmup': 50, 'num_sample': 50})
elif issubclass(estimator_type, PyroEstimator):
args.update({'num_steps': 10})
expected_num_parameters = 10
if seasonality == 52:
expected_num_parameters += 2
lgt = LGTFull(**args)
lgt.fit(train_df)
init_call = lgt.get_init_values()
if seasonality:
assert isinstance(init_call, LGTInitializer)
assert init_call.s == 52
init_values = init_call()
assert init_values['init_sea'].shape == (51, )
else:
assert not init_call
predict_df = lgt.predict(test_df)
expected_columns = ['week', 'prediction_5', 'prediction', 'prediction_95']
expected_shape = (51, len(expected_columns))
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert len(lgt._posterior_samples) == expected_num_parameters
def test_lgt_full_univariate_pyro(synthetic_data):
train_df, test_df, coef = synthetic_data
lgt = LGTFull(response_col='response',
date_col='week',
prediction_percentiles=[5, 95],
seasonality=52,
num_steps=10,
verbose=False,
estimator_type=PyroEstimatorVI)
lgt.fit(train_df)
predict_df = lgt.predict(test_df)
expected_columns = [
'week', 'prediction_lower', 'prediction', 'prediction_upper'
]
expected_shape = (51, len(expected_columns))
expected_num_parameters = 12 # no `lp__` in pyro
assert predict_df.shape == expected_shape
assert predict_df.columns.tolist() == expected_columns
assert len(lgt._posterior_samples) == expected_num_parameters