def test_fit_model_via_design_matrix_stats_models():
"""Testing the model fits via statsmodels module"""
df = generate_test_data_for_fitting(n=50, seed=41,
heteroscedastic=False)["df"]
df["y"] = df["y"].abs() + 1
model_formula_str = "y ~ x1_categ + x2 + x3"
design_mat_info = design_mat_from_formula(df,
model_formula_str,
pred_cols=None,
y_col=None)
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="statsmodels_ols")
expected = [14.0, 1.0, -3.1, -2.1, -0.3, 2.3, 0.7]
assert list(round(ml_model.params, 1).values) == expected
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="statsmodels_wls")
assert list(round(ml_model.params, 1).values) == expected
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="statsmodels_gls")
assert list(round(ml_model.params, 1).values) == expected
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="statsmodels_glm")
assert list(round(ml_model.params, 1).values) == [0.1, 0, 0, 0, 0, 0, 0]
def test_fit_model_via_design_matrix_error(design_mat_info):
"""Tests fit_model_via_design_matrix with """
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
with pytest.raises(ValueError,
match="The fit algorithm requested was not found"):
fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="unknown_model")
def test_fit_model_via_design_matrix_with_weights(data_with_weights):
df = data_with_weights["df"]
design_mat_info = data_with_weights["design_mat_info"]
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
sample_weight = df["w"]
# Ridge without weights
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="ridge",
sample_weight=None)
assert np.round(ml_model.intercept_, 0) == 15.0
assert ml_model.coef_[0].round() == 0.0
assert ml_model.coef_[1].round() == 0.0
# Ridge with weights
# Here we expect to get the coeffcients from: ``y[(n//2):] = 20 + -20 * x2[(n//2):]``
# This is becauase the weights are zero in the first half
# Therefore only the second equation (given above) will be relevant
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="ridge",
sample_weight=sample_weight)
"""
# commented out graphical test
# we expect to see two trends for y w.r.t x2
import plotly
from plotly import graph_objs as go
trace = go.Scatter(
x=df['x2'].values,
y=df['y'].values,
mode='markers')
data = [trace]
fig = go.Figure(data)
plotly.io.show(fig)
"""
assert np.round(ml_model.intercept_, 0) == 20.0
assert ml_model.coef_[0].round() == 0.0
assert ml_model.coef_[1].round() == -2.0
def test_fit_model_via_design_matrix2(design_mat_info):
"""Tests fit_model_via_design_matrix with elastic_net algorithm
and fit_algorithm_params"""
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="elastic_net",
fit_algorithm_params=dict(
n_alphas=100, eps=1e-2))
assert ml_model.coef_[0].round() == 0
assert ml_model.n_alphas == 100
assert ml_model.eps == 1e-2
assert ml_model.cv == 5 # from default parameters
def test_fit_model_via_design_matrix3(design_mat_info):
"""Tests fit_model_via_design_matrix with
elastic_net fit_algorithm and fit_algorithm_params"""
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="lasso_lars",
fit_algorithm_params=dict(
max_n_alphas=100,
eps=1e-2,
cv=2))
assert ml_model.coef_[0].round() == 0
assert ml_model.max_n_alphas == 100
assert ml_model.eps == 1e-2
assert ml_model.cv == 2 # override default
def test_fit_model_via_design_matrix(design_mat_info):
"""Tests fit_model_via_design_matrix"""
x_train = design_mat_info["x_mat"]
y_train = design_mat_info["y"]
sample_weight = np.array([1, 2, 3, 4, 5])
# Linear
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="linear")
assert ml_model.coef_[0].round() == 10.0
assert np.round(ml_model.intercept_, 1) == 0.0
# Ridge without weights
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="ridge",
sample_weight=None)
assert ml_model.coef_[0].round() == 0.0
assert np.round(ml_model.intercept_, 1) == 10.0
# Ridge with weights
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="ridge",
sample_weight=sample_weight)
assert ml_model.coef_[0].round() == 0.0
assert np.round(ml_model.intercept_, 1) == 10.0
# statsmodels_wls with weights
ml_model = fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="statsmodels_wls",
sample_weight=sample_weight)
assert ml_model.coef_[0].round() == 10.0
assert np.round(ml_model.intercept_, 1) == 0.0
with pytest.raises(ValueError, match="sample weights are passed."):
fit_model_via_design_matrix(x_train=x_train,
y_train=y_train,
fit_algorithm="lasso",
sample_weight=sample_weight)