def test_preprocess_data_scaling_std(self):
"""Test that the input features have been correctly scaled to 0.5 standard deviation."""
model = BPLModel(TEST_DATA, X=TEST_FEATS)
stan_data = model._pre_process_data()
self.assertTrue(
np.allclose(stan_data["X"].std(axis=0),
0.5 * np.ones(stan_data["X"].shape[1])))
def test_preprocess_data_scaling_mean(self):
"""Test that the input features have been correctly scaled to zero mean."""
model = BPLModel(TEST_DATA, X=TEST_FEATS)
stan_data = model._pre_process_data()
self.assertTrue(
np.allclose(stan_data["X"].mean(axis=0), np.zeros(stan_data["X"].shape[1]))
)
def test_preprocess_data_nofeats(self):
"""Test that stan data dictionary has the correct keys when no features are passed."""
model = BPLModel(TEST_DATA, X=None)
stan_data = model._pre_process_data()
self.assertTrue(
set(stan_data.keys())
== {"nteam", "nmatch", "home_team", "away_team", "home_goals", "away_goals"}
)
def test_preprocess_data_date(self):
"""Test that the correct date constraint is applied in preprocessing."""
model = BPLModel(TEST_DATA, X=TEST_FEATS)
stan_data = model._pre_process_data(max_date="2018-03-01")
df = TEST_DATA.copy()
df.loc[:, "date"] = pd.to_datetime(df["date"])
df = df[df["date"] <= "2018-03-01"]
n = len(df)
self.assertEqual(stan_data["nmatch"], n)
def test_fit_nofeats_summary(self):
"""Test that the summary is returned as a pandas dataframe when asked for."""
model = BPLModel(TEST_DATA)
summary = model.fit(return_summary=True, iter=100, seed=42)
self.assertIsInstance(summary, pd.DataFrame)
def test_fit_feats_run(self):
"""Test that the fit method runs when we pass features."""
model = BPLModel(TEST_DATA, X=TEST_FEATS)
model.fit(iter=100, seed=42)
def test_fit_nofeats_run(self):
"""Test that the fit method runs."""
model = BPLModel(TEST_DATA)
model.fit(iter=100, seed=42)
def test_preprocess_data_baddate(self):
"""Test that the correct error is raised if max_date is too low."""
model = BPLModel(TEST_DATA, X=TEST_FEATS)
self.assertRaises(ValueError, model._pre_process_data, max_date="1966-07-30")
def test_add_new_team(self):
"""Test functionality for adding new teams"""
model = BPLModel(data=TEST_DATA)
model_X = BPLModel(data=TEST_DATA, X=TEST_FEATS)
model.fit(iter=100)
model_X.fit(iter=100)
# check correct exception raised if unseen team is passed
self.assertRaises(
UnseenTeamError, model.overall_probabilities, "blah", "Man City"
)
self.assertRaises(
UnseenTeamError, model.overall_probabilities, "Man City", "blah"
)
self.assertRaises(
UnseenTeamError, model.overall_probabilities, "blah1", "blah2"
)
# add a new team and check shapes
model.add_new_team("blah1")
self.assertIn("blah1", model.team_indices.keys())
self.assertEqual(model.a.shape[1], 21)
self.assertEqual(model.b.shape[1], 21)
self.assertEqual(model.gamma.shape[1], 21)
# two new teams added with no extra prior information should be identical
model.add_new_team("blah2")
self.assertTupleEqual(
model.overall_probabilities("blah1", "blah2"),
model.overall_probabilities("blah2", "blah1"),
)
# if team already known is added, exception should be raised
self.assertRaises(ValueError, model.add_new_team, "Man City")
# test adding a new team to model with / without covariates
model_X.add_new_team("blah1")
model_X.add_new_team("blah2", X=np.array([70.0, 70.0, 70.0]))
import os
from unittest import TestCase
from bpl.models import BPLModel
from bpl.util import (
ModelNotConvergedWarning,
ModelNotFitError,
UnseenTeamError,
check_fit,
)
TEST_DATA = pd.read_csv(os.path.join(os.path.dirname(__file__), "test_data.csv"))
TEST_FEATS = pd.read_csv(os.path.join(os.path.dirname(__file__), "test_feats.csv"))
FITTED_MODEL = BPLModel(TEST_DATA)
FITTED_MODEL.fit(iter=1000, seed=42)
class TestBPLModel(TestCase):
def test_preprocess_data_nofeats(self):
"""Test that stan data dictionary has the correct keys when no features are passed."""
model = BPLModel(TEST_DATA, X=None)
stan_data = model._pre_process_data()
self.assertTrue(
set(stan_data.keys())
== {"nteam", "nmatch", "home_team", "away_team", "home_goals", "away_goals"}
)
def test_preprocess_data_feats(self):
"""Test that the stan data dictionary has the correct keys when features are passed."""