def test_causal_cto_w_seasons(date_rand_data, pre_str_period, post_str_period):
ci = CausalImpact(date_rand_data,
pre_str_period,
post_str_period,
nseasons=[{
'period': 4
}, {
'period': 3
}])
assert ci.model.freq_seasonal_periods == [4, 3]
assert ci.model.freq_seasonal_harmonics == [2, 1]
ci = CausalImpact(date_rand_data,
pre_str_period,
post_str_period,
nseasons=[{
'period': 4,
'harmonics': 1
}, {
'period': 3,
'harmonis': 1
}])
assert ci.model.freq_seasonal_periods == [4, 3]
assert ci.model.freq_seasonal_harmonics == [1, 1]
assert ci.inferences.index.dtype == date_rand_data.index.dtype
def test_other_formats(self):
# Test other data formats
pre_period = [1, 100]
post_period = [101, 200]
model_args = {"niter": 100}
# labelled dataframe
data = pd.DataFrame(np.random.randn(200, 3), columns=["a", "b", "c"])
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
actual_columns = impact.inferences.columns
assert actual_columns[0] == "response"
# numpy array
data = np.random.randn(200, 3)
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
actual_columns = impact.inferences.columns
assert actual_columns[0] == "response"
# numpy array
data = np.random.randn(200, 3)
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
actual_columns = impact.inferences.columns
assert actual_columns[0] == "response"
# list of lists
data = [[n, n + 2] for n in range(200)]
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
actual_columns = impact.inferences.columns
assert actual_columns[0] == "response"
def test_custom_model_input_validation(rand_data, pre_int_period,
post_int_period):
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, model='test')
assert str(
excinfo.value) == 'Input model must be of type UnobservedComponents.'
ucm = UnobservedComponents(rand_data.iloc[:101, 0],
level='llevel',
exog=rand_data.iloc[:101, 1:])
ucm.level = False
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, model=ucm)
assert str(excinfo.value) == 'Model must have level attribute set.'
ucm = UnobservedComponents(rand_data.iloc[:101, 0],
level='llevel',
exog=rand_data.iloc[:101, 1:])
ucm.exog = None
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, model=ucm)
assert str(excinfo.value) == 'Model must have exog attribute set.'
ucm = UnobservedComponents(rand_data.iloc[:101, 0],
level='llevel',
exog=rand_data.iloc[:101, 1:])
ucm.data = None
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, model=ucm)
assert str(excinfo.value) == 'Model must have data attribute set.'
def test_invalid_alpha_raises(rand_data, pre_int_period, post_int_period):
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, alpha=1)
assert str(excinfo.value) == 'alpha must be of type float.'
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, post_int_period, alpha=2.)
assert str(excinfo.value) == (
'alpha must range between 0 (zero) and 1 (one) inclusive.')
def test_causal_cto_raises_on_None_input(rand_data, pre_int_period, post_int_period):
with pytest.raises(ValueError) as excinfo:
CausalImpact(None, pre_int_period, post_int_period)
assert str(excinfo.value) == 'data input cannot be empty'
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, None, post_int_period)
assert str(excinfo.value) == 'pre_period input cannot be empty'
with pytest.raises(ValueError) as excinfo:
CausalImpact(rand_data, pre_int_period, None)
assert str(excinfo.value) == 'post_period input cannot be empty'
def test_invalid_data_input_raises():
with pytest.raises(ValueError) as excinfo:
CausalImpact('test', [0, 5], [5, 10])
assert str(excinfo.value) == 'Could not transform input data to pandas DataFrame.'
data = [1, 2, 3, 4, 5, 6, 2 + 1j]
with pytest.raises(ValueError) as excinfo:
CausalImpact(data, [0, 3], [3, 6])
assert str(excinfo.value) == 'Input data must contain only numeric values.'
data = np.random.randn(10, 2)
data[0, 1] = np.nan
with pytest.raises(ValueError) as excinfo:
CausalImpact(data, [0, 3], [3, 6])
assert str(excinfo.value) == 'Input data cannot have NAN values.'
def test_kwargs_validation(rand_data, pre_int_period, post_int_period):
with pytest.raises(ValueError) as excinfo:
ci = CausalImpact(rand_data,
pre_int_period,
post_int_period,
standardize='yes')
assert str(excinfo.value) == 'Standardize argument must be of type bool.'
def test_default_causal_cto(rand_data, pre_int_period, post_int_period):
ci = CausalImpact(rand_data, pre_int_period, post_int_period)
assert_frame_equal(ci.data, rand_data)
assert ci.pre_period == pre_int_period
assert ci.post_period == post_int_period
pre_data = rand_data.loc[pre_int_period[0]:pre_int_period[1], :]
assert_frame_equal(ci.pre_data, pre_data)
post_data = rand_data.loc[post_int_period[0]:post_int_period[1], :]
assert_frame_equal(ci.post_data, post_data)
assert ci.alpha == 0.05
normed_pre_data, (mu, sig) = standardize(pre_data)
assert_frame_equal(ci.normed_pre_data, normed_pre_data)
normed_post_data = (post_data - mu) / sig
assert_frame_equal(ci.normed_post_data, normed_post_data)
assert ci.mu_sig == (mu[0], sig[0])
assert ci.model_args == {
'fit_method': 'hmc',
'niter': 1000,
'prior_level_sd': 0.01,
'season_duration': 1,
'nseasons': 1,
'standardize': True
}
assert isinstance(ci.model, tfp.sts.Sum)
design_matrix = ci.model.components[1].design_matrix.to_dense()
assert_array_equal(
design_matrix,
pd.concat([normed_pre_data,
normed_post_data]).astype(np.float32).iloc[:, 1:])
assert ci.inferences is not None
assert ci.inferences.index.dtype == rand_data.index.dtype
assert ci.summary_data is not None
assert ci.p_value > 0 and ci.p_value < 1
assert ci.model_args['niter'] == 1000
assert ci.model_samples is not None
def test_summarizer(monkeypatch, rand_data, pre_int_period, post_int_period):
summarizer_mock = mock.Mock()
fit_mock = mock.Mock()
process_mock = mock.Mock()
summarize_mock = mock.Mock()
monkeypatch.setattr('causalimpact.main.CausalImpact._fit_model', fit_mock)
monkeypatch.setattr('causalimpact.main.CausalImpact._summarize_inferences',
summarize_mock)
monkeypatch.setattr(
'causalimpact.main.CausalImpact._process_posterior_inferences',
process_mock)
monkeypatch.setattr('causalimpact.main.summarizer', summarizer_mock)
ci = CausalImpact(rand_data,
pre_int_period,
post_int_period,
model_args={'fit_method': 'vi'})
ci.summary_data = 'summary_data'
ci.p_value = 0.5
ci.alpha = 0.05
ci.summary()
summarizer_mock.summary.assert_called_with('summary_data', 0.5, 0.05,
'summary', 2)
with pytest.raises(ValueError) as excinfo:
ci.summary(digits='1')
assert str(
excinfo.value) == ('Input value for digits must be integer. Received '
'"<class \'str\'>" instead.')
def test_plotter(monkeypatch, rand_data, pre_int_period, post_int_period):
plotter_mock = mock.Mock()
fit_mock = mock.Mock()
process_mock = mock.Mock()
summarize_mock = mock.Mock()
monkeypatch.setattr('causalimpact.main.CausalImpact._fit_model', fit_mock)
monkeypatch.setattr('causalimpact.main.CausalImpact._summarize_inferences',
summarize_mock)
monkeypatch.setattr(
'causalimpact.main.CausalImpact._process_posterior_inferences',
process_mock)
monkeypatch.setattr('causalimpact.main.plotter', plotter_mock)
ci = CausalImpact(rand_data,
pre_int_period,
post_int_period,
model_args={'fit_method': 'vi'})
ci.inferences = 'inferences'
ci.pre_data = 'pre_data'
ci.post_data = 'post_data'
ci.plot()
plotter_mock.plot.assert_called_with(
'inferences',
'pre_data',
'post_data',
panels=['original', 'pointwise', 'cumulative'],
figsize=(10, 7))
def get_results_from_statsmodels(
processed_data: pd.DataFrame,
experiment_config: ExperimentConfig) -> ExperimentOutput:
"""
Get results from the CausalImpact implementation that uses statsmodels
as a backend
"""
if PYTHON_IMPLEMENTATION_USED != "pycausalimpact":
raise EnvironmentError(
f"tfcausalimpact not currently used. Implementations: {PYTHON_IMPLEMENTATION_USED}"
)
ci = CausalImpact(
processed_data,
experiment_config.training_period,
experiment_config.evaluation_period,
)
experiment_output = ExperimentOutput(
**{
"results_summary": ci.summary_data,
"trained_model": ci,
"experiment_name": experiment_config.experiment_name,
})
return experiment_output
def test_gap(self):
post_period = [120, 200]
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
assert np.all(
pd.isnull(impact.inferences.loc[101:119,
impact.inferences.columns[2:]]))
def test_causal_cto_with_no_standardization(rand_data, pre_int_period, post_int_period):
ci = CausalImpact(rand_data, pre_int_period, post_int_period, model_args=dict(
standardize=False, fit_method='vi'))
assert ci.normed_pre_data is None
assert ci.normed_post_data is None
assert ci.mu_sig is None
assert ci.p_value > 0 and ci.p_value < 1
def test_summary_w_report_output(self, monkeypatch, inference_input,
summary_report_filename):
inferences_df = pd.DataFrame(inference_input)
causal = CausalImpact()
params = {'alpha': 0.05, 'post_period': [2, 4]}
causal.params = params
causal.inferences = inferences_df
dedent_mock = mock.Mock()
expected = open(summary_report_filename).read()
expected = re.sub(r'\s+', ' ', expected)
expected = expected.strip()
tmpdir = mkdtemp()
tmp_file = os.path.join(tmpdir, 'summary_test')
def dedent_side_effect(msg):
with open(tmp_file, 'a') as file_obj:
msg = re.sub(r'\s+', ' ', msg)
msg = msg.strip()
file_obj.write(msg)
return msg
dedent_mock.side_effect = dedent_side_effect
monkeypatch.setattr('textwrap.dedent', dedent_mock)
causal.summary(output='report')
result_str = open(tmp_file, 'r').read()
assert result_str == expected
def test_missing_values_in_pre_period_y(self, pre_period, post_period):
data = pd.DataFrame(np.random.randn(200, 3), columns=["y", "x1", "x2"])
data.iloc[95:100, 0] = np.nan
impact = CausalImpact(data, pre_period, post_period)
impact.run()
"""Test that all columns in the result series except those associated
with point predictions have missing values at the time points the
result time series has missing values."""
predicted_cols = [
impact.inferences.columns.get_loc(col)
for col in impact.inferences.columns
if ("response" not in col and "point_effect" not in col)
]
effect_cols = [
impact.inferences.columns.get_loc(col)
for col in impact.inferences.columns if "point_effect" in col
]
response_cols = [
impact.inferences.columns.get_loc(col)
for col in impact.inferences.columns if "response" in col
]
assert np.all(np.isnan(impact.inferences.iloc[95:100, response_cols]))
assert (np.any(np.isnan(
impact.inferences.iloc[95:100, predicted_cols])) == False)
assert np.any(np.isnan(impact.inferences.iloc[:95, :])) == False
assert np.any(np.isnan(impact.inferences.iloc[101:, :])) == False
def test_gap_between_pre_and_post_periods(self, data, pre_period):
post_period = [120, 199]
impact = CausalImpact(data, pre_period, post_period)
impact.run()
assert np.all(
pd.isnull(impact.inferences.loc[101:119,
impact.inferences.columns[2:]]))
def test_post_period_finishes_before_end_of_data(self, data, pre_period):
post_period = [101, 197]
impact = CausalImpact(data, pre_period, post_period)
impact.run()
np.testing.assert_array_equal(impact.inferences.response.values,
data.y.values)
assert np.all(pd.isnull(impact.inferences.iloc[-2:, 2:]))
def test_pre_period_starts_after_beginning_of_data(self, data):
pre_period = [3, 100]
impact = CausalImpact(data, pre_period, [101, 199])
impact.run()
np.testing.assert_array_equal(impact.inferences.response.values,
data.y.values)
assert np.all(pd.isnull(impact.inferences.iloc[0:pre_period[0], 2:]))
def get_results_from_tensorflow(
processed_data: pd.DataFrame,
experiment_config: ExperimentConfig) -> ExperimentOutput:
if PYTHON_IMPLEMENTATION_USED != "tfcausalimpact":
raise EnvironmentError(
f"tfcausalimpact not currently used. Implementations: {PYTHON_IMPLEMENTATION_USED}"
)
causal_model_config = experiment_config.causal_model_config
alpha, model, model_args = process_causal_impact_config(
causal_model_config)
print("things passed to the causal model: ")
print(alpha, model, model_args)
ci = CausalImpact(
processed_data,
experiment_config.training_period,
experiment_config.evaluation_period,
alpha=alpha,
model=model,
model_args=model_args,
)
experiment_output = ExperimentOutput(
**{
"results_summary": ci.summary_data,
"trained_model": ci,
"experiment_name": experiment_config.experiment_name,
})
return experiment_output
def test_default_causal_inferences(fix_path):
np.random.seed(1)
data = pd.read_csv(os.path.join(fix_path, 'google_data.csv'))
del data['t']
pre_period = [0, 60]
post_period = [61, 90]
ci = CausalImpact(data, pre_period, post_period)
assert int(ci.summary_data['average']['actual']) == 156
assert int(ci.summary_data['average']['predicted']) == 129
assert int(ci.summary_data['average']['predicted_lower']) == 102
assert int(ci.summary_data['average']['predicted_upper']) == 156
assert int(ci.summary_data['average']['abs_effect']) == 26
assert round(ci.summary_data['average']['abs_effect_lower'], 1) == -0.2
assert int(ci.summary_data['average']['abs_effect_upper']) == 53
assert round(ci.summary_data['average']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['average']['rel_effect_lower'], 1) == 0.0
assert round(ci.summary_data['average']['rel_effect_upper'], 1) == 0.4
assert int(ci.summary_data['cumulative']['actual']) == 4687
assert int(ci.summary_data['cumulative']['predicted']) == 3883
assert int(ci.summary_data['cumulative']['predicted_lower']) == 3085
assert int(ci.summary_data['cumulative']['predicted_upper']) == 4693
assert int(ci.summary_data['cumulative']['abs_effect']) == 803
assert round(ci.summary_data['cumulative']['abs_effect_lower'], 1) == -6.8
assert int(ci.summary_data['cumulative']['abs_effect_upper']) == 1601
assert round(ci.summary_data['cumulative']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['cumulative']['rel_effect_lower'], 1) == 0.0
assert round(ci.summary_data['cumulative']['rel_effect_upper'], 1) == 0.4
assert round(ci.p_value, 1) == 0.0
def test_default_causal_inferences_w_date(fix_path):
np.random.seed(1)
data = pd.read_csv(os.path.join(fix_path, 'google_data.csv'))
data['date'] = pd.to_datetime(data['t'])
data.index = data['date']
del data['t']
del data['date']
pre_period = ['2016-02-20 22:41:20', '2016-02-20 22:51:20']
post_period = ['2016-02-20 22:51:30', '2016-02-20 22:56:20']
ci = CausalImpact(data, pre_period, post_period)
assert int(ci.summary_data['average']['actual']) == 156
assert int(ci.summary_data['average']['predicted']) == 129
assert int(ci.summary_data['average']['predicted_lower']) == 102
assert int(ci.summary_data['average']['predicted_upper']) == 156
assert int(ci.summary_data['average']['abs_effect']) == 26
assert round(ci.summary_data['average']['abs_effect_lower'], 1) == -0.2
assert int(ci.summary_data['average']['abs_effect_upper']) == 53
assert round(ci.summary_data['average']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['average']['rel_effect_lower'], 1) == 0.0
assert round(ci.summary_data['average']['rel_effect_upper'], 1) == 0.4
assert int(ci.summary_data['cumulative']['actual']) == 4687
assert int(ci.summary_data['cumulative']['predicted']) == 3883
assert int(ci.summary_data['cumulative']['predicted_lower']) == 3085
assert int(ci.summary_data['cumulative']['predicted_upper']) == 4693
assert int(ci.summary_data['cumulative']['abs_effect']) == 803
assert round(ci.summary_data['cumulative']['abs_effect_lower'], 1) == -6.8
assert int(ci.summary_data['cumulative']['abs_effect_upper']) == 1601
assert round(ci.summary_data['cumulative']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['cumulative']['rel_effect_lower'], 1) == 0.0
assert round(ci.summary_data['cumulative']['rel_effect_upper'], 1) == 0.4
assert round(ci.p_value, 1) == 0.0
def test_string_index_with_no_date_formatted(rand_data, pre_int_period,
post_int_period):
rand_data.set_index(rand_data.index.map(str), inplace=True)
pre_period = ['0', '60']
post_period = ['61', '90']
_ = CausalImpact(rand_data, pre_period, post_period)
def test_plot_cumulative_panel(rand_data, pre_int_period, post_int_period, monkeypatch):
ci = CausalImpact(rand_data, pre_int_period, post_int_period)
ax_mock = mock.Mock()
plotter_mock = mock.Mock()
plotter_mock.subplot.return_value = ax_mock
plot_mock = mock.Mock(return_value=plotter_mock)
monkeypatch.setattr(plot.Plot, '_get_plotter', plot_mock)
ci.plot(panels=['cumulative'])
plot_mock.assert_called_once()
plotter_mock.figure.assert_called_with(figsize=(15, 12))
plotter_mock.subplot.assert_any_call(1, 1, 1, sharex=ax_mock)
ax_args = ax_mock.plot.call_args
inferences = ci.inferences.iloc[1:, :]
assert_array_equal(inferences['post_cum_effects'], ax_args[0][0])
assert ax_args[0][1] == 'b--'
assert ax_args[1] == {'label': 'Cumulative Effect'}
ax_mock.axvline.assert_called_with(ci.post_period[0] - 1, c='k', linestyle='--')
ax_args = ax_mock.fill_between.call_args_list[0]
assert_array_equal(ax_args[0][0], inferences['post_cum_effects'].index)
assert_array_equal(ax_args[0][1], inferences['post_cum_effects_lower'])
assert_array_equal(ax_args[0][2], inferences['post_cum_effects_upper'])
assert ax_args[1] == {'facecolor': 'blue', 'interpolate': True, 'alpha': 0.25}
ax_mock.axhline.assert_called_with(y=0, color='k', linestyle='--')
ax_mock.grid.assert_called_with(True, linestyle='--')
ax_mock.legend.assert_called()
plotter_mock.show.assert_called_once()
def test_default_causal_inferences_w_str_date(fix_path):
np.random.seed(1)
data = pd.read_csv(os.path.join(fix_path, 'volks_data.csv'),
header=0,
sep=' ',
index_col='Date')
pre_period = [np.min(data.index.values), pd.Timestamp('2015-09-13')]
post_period = [pd.Timestamp('2015-09-20'), np.max(data.index.values)]
ci = CausalImpact(data, pre_period, post_period)
assert int(ci.summary_data['average']['actual']) == 126
assert int(ci.summary_data['average']['predicted']) == 171
assert int(ci.summary_data['average']['predicted_lower']) == 165
assert int(ci.summary_data['average']['predicted_upper']) == 177
assert int(ci.summary_data['average']['abs_effect']) == -44
assert round(ci.summary_data['average']['abs_effect_lower'], 1) == -50.4
assert int(ci.summary_data['average']['abs_effect_upper']) == -39
assert round(ci.summary_data['average']['rel_effect'], 1) == -0.3
assert round(ci.summary_data['average']['rel_effect_lower'], 2) == -0.29
assert round(ci.summary_data['average']['rel_effect_upper'], 2) == -0.23
assert int(ci.summary_data['cumulative']['actual']) == 10026
assert int(ci.summary_data['cumulative']['predicted']) == 13574
assert int(ci.summary_data['cumulative']['predicted_lower']) == 13113
assert int(ci.summary_data['cumulative']['predicted_upper']) == 14004
assert int(ci.summary_data['cumulative']['abs_effect']) == -3548
assert int(ci.summary_data['cumulative']['abs_effect_lower']) == -3977
assert int(ci.summary_data['cumulative']['abs_effect_upper']) == -3087
assert round(ci.summary_data['cumulative']['rel_effect'], 1) == -0.3
assert round(ci.summary_data['cumulative']['rel_effect_lower'], 2) == -0.29
assert round(ci.summary_data['cumulative']['rel_effect_upper'], 2) == -0.23
assert round(ci.p_value, 1) == 0.0
assert ci.inferences.index.dtype == data.index.dtype
def test_early_post_period(self):
post_period = [101, 197]
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
assert pd.util.testing.assert_numpy_array_equal(
impact.inferences.response.values, data.y.values)
assert np.all(pd.isnull(impact.inferences.iloc[-2:, 2:]))
def test_default_model_sparse_linear_regression_arma_data():
data = pd.read_csv('tests/fixtures/arma_sparse_reg.csv')
data.iloc[70:, 0] += 5
pre_period = [0, 69]
post_period = [70, 99]
ci = CausalImpact(data, pre_period, post_period)
samples = ci.model_samples
# Weights are computed as per original TFP source code:
# https://github.com/tensorflow/probability/blob/v0.12.1/tensorflow_probability/python/sts/regression.py#L489-L494 # noqa: E501
global_scale = (
samples['SparseLinearRegression/_global_scale_noncentered'] *
tf.sqrt(samples['SparseLinearRegression/_global_scale_variance']) * 0.1
)
local_scales = (
samples['SparseLinearRegression/_local_scales_noncentered'] *
tf.sqrt(samples['SparseLinearRegression/_local_scale_variances'])
)
weights = (
samples['SparseLinearRegression/_weights_noncentered'] * local_scales *
global_scale[..., tf.newaxis]
)
assert tf.abs(tf.reduce_mean(weights, axis=0).numpy()[1]) < 0.05
def test_default_causal_inferences(fix_path):
np.random.seed(1)
data = pd.read_csv(os.path.join(fix_path, 'google_data.csv'))
del data['t']
pre_period = [0, 60]
post_period = [61, 90]
ci = CausalImpact(data, pre_period, post_period)
assert int(ci.summary_data['average']['actual']) == 156
assert int(ci.summary_data['average']['predicted']) == 129
assert int(ci.summary_data['average']['predicted_lower']) == 124
assert int(ci.summary_data['average']['predicted_upper']) == 134
assert int(ci.summary_data['average']['abs_effect']) == 27
assert round(ci.summary_data['average']['abs_effect_lower'], 1) == 21.6
assert int(ci.summary_data['average']['abs_effect_upper']) == 31
assert round(ci.summary_data['average']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['average']['rel_effect_lower'], 2) == 0.17
assert round(ci.summary_data['average']['rel_effect_upper'], 2) == 0.25
assert int(ci.summary_data['cumulative']['actual']) == 4687
assert int(ci.summary_data['cumulative']['predicted']) == 3876
assert int(ci.summary_data['cumulative']['predicted_lower']) == 3729
assert int(ci.summary_data['cumulative']['predicted_upper']) == 4040
assert int(ci.summary_data['cumulative']['abs_effect']) == 810
assert int(ci.summary_data['cumulative']['abs_effect_lower']) == 646
assert int(ci.summary_data['cumulative']['abs_effect_upper']) == 957
assert round(ci.summary_data['cumulative']['rel_effect'], 1) == 0.2
assert round(ci.summary_data['cumulative']['rel_effect_lower'], 2) == 0.17
assert round(ci.summary_data['cumulative']['rel_effect_upper'], 2) == 0.25
assert round(ci.p_value, 1) == 0.0
assert ci.inferences.index.dtype == data.index.dtype
def test_default_causal_cto_no_covariates(rand_data, pre_int_period,
post_int_period):
rand_data = pd.DataFrame(rand_data.iloc[:, 0])
ci = CausalImpact(rand_data, pre_int_period, post_int_period)
assert_frame_equal(ci.data, rand_data)
assert ci.pre_period == pre_int_period
assert ci.post_period == post_int_period
pre_data = rand_data.loc[pre_int_period[0]:pre_int_period[1], :]
assert_frame_equal(ci.pre_data, pre_data)
post_data = rand_data.loc[post_int_period[0]:post_int_period[1], :]
assert_frame_equal(ci.post_data, post_data)
assert ci.alpha == 0.05
normed_pre_data, (mu, sig) = standardize(pre_data)
assert_frame_equal(ci.normed_pre_data, normed_pre_data)
normed_post_data = (post_data - mu) / sig
assert_frame_equal(ci.normed_post_data, normed_post_data)
assert ci.mu_sig == (mu[0], sig[0])
assert ci.model_args == {
'fit_method': 'hmc',
'niter': 1000,
'prior_level_sd': 0.01,
'season_duration': 1,
'nseasons': 1,
'standardize': True
}
assert isinstance(ci.model, tfp.sts.LocalLevel)
assert ci.inferences is not None
assert ci.inferences.index.dtype == rand_data.index.dtype
assert ci.summary_data is not None
assert ci.p_value > 0 and ci.p_value < 1
assert ci.model_args['niter'] == 1000
assert ci.model_samples is not None
def main():
with open("/path/to/glowroot/data.json", 'rt') as file:
glowroot_data = json.load(file)
x = list()
y = list()
for point in glowroot_data["dataSeries"][0]["data"]:
x.append(point[0])
y.append(point[1])
data_frame = pd.DataFrame({
"timestamp": pd.to_datetime(x, unit="ms"),
"y": y
})
data_frame = data_frame.set_index("timestamp")
logging.info(data_frame)
pre_period = [
pd.to_datetime(1573661277259, unit="ms"),
pd.to_datetime(1573661647328, unit="ms")
]
post_period = [
pd.to_datetime(1573661652328, unit="ms"),
pd.to_datetime(1573661932369, unit="ms")
]
causal_impact = CausalImpact(data_frame,
pre_period,
post_period,
prior_level_sd=0.1)
logging.info(causal_impact.summary())
causal_impact.plot()
def test_late_pre_period(self):
pre_period = [3, 100]
impact = CausalImpact(data, pre_period, post_period, model_args)
impact.run()
assert pd.util.testing.assert_numpy_array_equal(
impact.inferences.response.values, data.y.values)
assert np.all(pd.isnull(impact.inferences.iloc[0:pre_period[0], 2:]))