def test_term_new_data_numeric():
data = pd.DataFrame({"x": [10, 10, 10]})
var_expr = Parser(Scanner("x").scan(False)).parse()
var_term = Variable(var_expr.name.lexeme, var_expr.level)
var_term.set_type(data)
var_term.set_data()
assert (var_term.value == [10, 10, 10]).all()
data = pd.DataFrame({"x": [1, 2, 3]})
assert (var_term.eval_new_data(data).T == [1, 2, 3]).all()
def test_term_new_data_categoric():
data = pd.DataFrame({"x": ["A", "B", "C"]})
# Full rank encoding
var_expr = Parser(Scanner("x").scan(False)).parse()
var_term = Variable(var_expr.name.lexeme, var_expr.level)
var_term.set_type(data)
var_term.set_data(spans_intercept=True)
assert (np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]]) == var_term.value).all()
data = pd.DataFrame({"x": ["B", "C"]})
assert (var_term.eval_new_data(data) == np.array([[0, 1, 0], [0, 0, 1]])).all()
# It remembers it saw "A", "B", and "C", but not "D".
# So when you pass a new level, it raises a ValueError.
with pytest.raises(
ValueError, match="The levels D in 'x' are not present in the original data set."
):
data = pd.DataFrame({"x": ["B", "C", "D"]})
var_term.eval_new_data(data)
# The same with reduced encoding
data = pd.DataFrame({"x": ["A", "B", "C"]})
var_expr = Parser(Scanner("x").scan(False)).parse()
var_term = Variable(var_expr.name.lexeme, var_expr.level)
var_term.set_type(data)
var_term.set_data()
assert (np.array([[0, 0], [1, 0], [0, 1]]) == var_term.value).all()
data = pd.DataFrame({"x": ["A", "C"]})
assert (var_term.eval_new_data(data) == np.array([[0, 0], [0, 1]])).all()
# It remembers it saw "A", "B", and "C", but not "D".
# So when you pass a new level, it raises a ValueError.
with pytest.raises(
ValueError, match="The levels D in 'x' are not present in the original data set."
):
data = pd.DataFrame({"x": ["B", "C", "D"]})
var_term.eval_new_data(data)
