-
Notifications
You must be signed in to change notification settings - Fork 1.7k
/
test_time_series_blending.py
178 lines (145 loc) · 6.61 KB
/
test_time_series_blending.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
"""Module to test time_series `blend_model` functionality
"""
import random
import numpy as np
import pandas as pd
import pytest
from pycaret.datasets import get_data
from pycaret.time_series import TSForecastingExperiment
##########################
# Tests Start Here ####
##########################
@pytest.mark.filterwarnings(
"ignore::statsmodels.tools.sm_exceptions.ConvergenceWarning:statsmodels"
)
@pytest.mark.parametrize("method", ["mean", "median", "min", "max", "gmean"])
def test_blend_model_basic(load_setup, load_models, method):
"""Tests basic blender functionality for all methods"""
from sktime.forecasting.compose import EnsembleForecaster
exp = load_setup
models = load_models
weights = [random.uniform(0, 1) for _ in range(len(models))]
blender = exp.blend_models(models, method=method, weights=weights, verbose=False)
assert isinstance(blender, EnsembleForecaster)
# Test input models are available
blender_forecasters = blender.forecasters_
blender_forecasters_class = [f.__class__ for f in blender_forecasters]
ts_models_class = [f.__class__ for f in models]
assert blender_forecasters_class == ts_models_class
def test_blend_models_tuning():
"""Test the tuning of blended models."""
data = get_data("airline", verbose=False)
exp = TSForecastingExperiment()
exp.setup(data=data, fh=12, fold=2, session_id=42)
model1 = exp.create_model("naive")
model2 = exp.create_model("ets")
model3 = exp.create_model("lr_cds_dt")
blender = exp.blend_models([model1, model2, model3])
_, tuner = exp.tune_model(blender, return_tuner=True)
assert len(pd.DataFrame(tuner.cv_results_)) > 1
@pytest.mark.filterwarnings(
"ignore::statsmodels.tools.sm_exceptions.ConvergenceWarning:statsmodels"
)
def test_blend_model_predict(load_setup, load_models):
"""Test to make sure that blending predictions are different when they need
to be and same when they need to be (depending on the hyperparameters).
"""
exp = load_setup
models = load_models
random.seed(42)
weights = [random.uniform(0, 1) for _ in range(len(models))]
# -------------------------------------------------------------------------#
# Prediction should be different for different methods
# -------------------------------------------------------------------------#
mean_blender = exp.blend_models(models, method="mean")
gmean_blender = exp.blend_models(models, method="gmean")
median_blender = exp.blend_models(models, method="median")
min_blender = exp.blend_models(models, method="min")
max_blender = exp.blend_models(models, method="max")
mean_blender_w_wts = exp.blend_models(models, method="mean", weights=weights)
gmean_blender_w_wts = exp.blend_models(models, method="gmean", weights=weights)
median_blender_w_wts = exp.blend_models(models, method="median", weights=weights)
min_blender_w_wts = exp.blend_models(models, method="min", weights=weights)
max_blender_w_wts = exp.blend_models(models, method="max", weights=weights)
mean_blender_pred = exp.predict_model(mean_blender)
gmean_blender_pred = exp.predict_model(gmean_blender)
median_blender_pred = exp.predict_model(median_blender)
min_blender_pred = exp.predict_model(min_blender)
max_blender_pred = exp.predict_model(max_blender)
mean_blender_w_wts_pred = exp.predict_model(mean_blender_w_wts)
gmean_blender_w_wts_pred = exp.predict_model(gmean_blender_w_wts)
median_blender_w_wts_pred = exp.predict_model(median_blender_w_wts)
min_blender_w_wts_pred = exp.predict_model(min_blender_w_wts)
max_blender_w_wts_pred = exp.predict_model(max_blender_w_wts)
different_preds = [
mean_blender_pred,
gmean_blender_pred,
median_blender_pred,
min_blender_pred,
max_blender_pred,
mean_blender_w_wts_pred,
gmean_blender_w_wts_pred,
median_blender_w_wts_pred,
]
for i, _ in enumerate(different_preds):
for j in range(i + 1, len(different_preds)):
assert not np.array_equal(different_preds[i], different_preds[j])
# -------------------------------------------------------------------------#
# Prediction for some methods should not be impacted by weights
# e.g. min, max
# -------------------------------------------------------------------------#
assert np.array_equal(
min_blender_pred, min_blender_w_wts_pred
), "min blender predictions with and without weights are not the same"
assert np.array_equal(
max_blender_pred, max_blender_w_wts_pred
), "max blender predictions with and without weights are not the same"
def test_blend_model_custom_folds(load_pos_and_neg_data):
"""Test custom folds in blend_model"""
exp = TSForecastingExperiment()
setup_fold = 3
exp.setup(
data=load_pos_and_neg_data,
fold=setup_fold,
fh=12,
fold_strategy="sliding",
verbose=False,
)
#######################################
# Test Tune Model with custom folds ##
#######################################
model = exp.create_model("naive")
_ = exp.blend_models([model, model, model])
metrics1 = exp.pull()
custom_fold = 5
_ = exp.blend_models([model, model, model], fold=custom_fold)
metrics2 = exp.pull()
assert len(metrics1) == setup_fold + 2 # + 2 for Mean and SD
assert len(metrics2) == custom_fold + 2 # + 2 for Mean and SD
def test_blend_with_larger_predict_fh():
"""Test to make sure that blending predictions work when the forecast horizon
used in predictions is larger than the one used for training
Ref: https://github.com/pycaret/pycaret/issues/2329
"""
data = get_data("airline", verbose=False)
exp = TSForecastingExperiment()
exp.setup(data=data, fh=12, fold=2, session_id=42)
model1 = exp.create_model("naive")
model2 = exp.create_model("ets")
model3 = exp.create_model("lr_cds_dt")
blender = exp.blend_models([model1, model2, model3])
# Check that forecasts can be created for FH greater than the one used for training
FHs = [12, 24]
for fh in FHs:
preds = exp.predict_model(blender, fh=fh)
assert len(preds) == fh
def test_error_conditions(load_setup, load_models):
"""Tests error conditions for blend_models"""
exp = load_setup
models = load_models
random.seed(42)
weights = [random.uniform(0, 1) for _ in range(len(models))]
with pytest.raises(ValueError) as err_msg:
_ = exp.blend_models(models, method="voting", weights=weights)
exception_msg = err_msg.value.args[0]
assert "method 'voting' is not supported from pycaret 3.0.1" in exception_msg