def test_k_fold_splitter():
# testing without stratification
result, logs = k_fold_splitter(sample_data, 2, random_state=42)
assert len(result) == 2
train_1_idx = result[0][0]
test_1_idx = result[0][1][0]
train_2_idx = result[1][0]
test_2_idx = result[1][1][0]
assert set(train_1_idx) == set(test_2_idx)
assert set(test_1_idx) == set(train_2_idx)
# testing with stratification
result, logs = k_fold_splitter(sample_data, 2, random_state=42, stratify_column='space')
assert len(result) == 2
train_1_idx = result[0][0]
test_1_idx = result[0][1][0]
train_2_idx = result[1][0]
test_2_idx = result[1][1][0]
assert set(train_1_idx) == set(test_2_idx)
assert set(test_1_idx) == set(train_2_idx)
train_1_strat = sample_data.iloc[train_1_idx]['space']
test_1_strat = sample_data.iloc[test_1_idx]['space']
train_2_strat = sample_data.iloc[train_2_idx]['space']
test_2_strat = sample_data.iloc[test_2_idx]['space']
assert train_1_strat.nunique() == test_2_strat.nunique()
assert train_2_strat.nunique() == test_1_strat.nunique()
from toolz import curry
from fklearn.tuning.parameter_tuners import grid_search_cv
from fklearn.training.regression import linear_regression_learner
from fklearn.validation.splitters import k_fold_splitter
from fklearn.validation.evaluators import mse_evaluator
space = {
"intercept": lambda: [0, 1],
}
@curry
def param_train_fn(space, train_set):
return linear_regression_learner(
features=["income"],
target="target",
params={"fit_intercept": space["intercept"]},
)(train_set)
split_fn = k_fold_splitter(n_splits=2)
tuning_log = grid_search_cv(space,
df,
param_train_fn=param_train_fn,
split_fn=split_fn,
eval_fn=mse_evaluator)
for run in tuning_log:
print(f"mse score in this run per fold: {run['validator_log']}")
print(f"with parameters: {run['iter_space']}")
def split_fn():
return k_fold_splitter(n_splits=2, random_state=30)