feat(inverse_transform): enable fit and transform with horizontal_mat…

…rix (#139)

* feat(inverse_transform): enable fit and transform with horizontal_matrix

* feat(inverse_transform): add docstring and feedbacks

saiph/inverse_transform.py

@@ -6,7 +6,6 @@
 import pandas as pd
 from numpy.typing import NDArray
 
-from saiph.exception import InvalidParameterException
 from saiph.models import Model
 from saiph.reduction import DUMMIES_SEPARATOR
 from saiph.reduction.utils.common import get_dummies_mapping
@@ -16,16 +15,13 @@ def inverse_transform(
     coord: pd.DataFrame,
     model: Model,
     *,
-    use_approximate_inverse: bool = False,
     use_max_modalities: bool = True,
 ) -> pd.DataFrame:
     """Return original format dataframe from coordinates.
 
     Parameters:
         coord: coord of individuals to reverse transform
         model: model used for projection
-        use_approximate_inverse: matrix is not invertible when n_individuals < n_dimensions
-            an approximation with bias can be done by setting to ``True``. default: ``False``
         use_max_modalities: for each variable, it assigns to the individual
             the modality with the highest proportion (True)
             or a random modality weighted by their proportion (False). default: True
@@ -36,16 +32,7 @@ def inverse_transform(
     """
     random_gen = np.random.default_rng(model.seed)
 
-    # Check dimension size regarding N
-    n_dimensions = len(model.dummy_categorical) + len(model.original_continuous)
-    n_records = len(coord)
-
-    if not use_approximate_inverse and n_records < n_dimensions:
-        raise InvalidParameterException(
-            f"n_dimensions ({n_dimensions}) is greater than n_records ({n_records})."
-        )
     # Get back scaled_values from coord with inverse matrix operation
-    # If n_records < n_dimensions, There will be an approximation of the inverse of V.T
     scaled_values = pd.DataFrame(coord @ np.linalg.pinv(model.V.T))
     # get number of continuous variables
     nb_quanti = len(model.original_continuous)

saiph/inverse_transform_test.py

@@ -6,7 +6,6 @@
 from numpy.testing import assert_allclose
 from pandas.testing import assert_frame_equal, assert_series_equal
 
-from saiph.exception import InvalidParameterException
 from saiph.inverse_transform import (
     get_random_weighted_columns,
     inverse_transform,
@@ -110,19 +109,6 @@ def test_undummify_when_dummies_prefix_is_in_variable_name() -> None:
     assert_frame_equal(df, expected)
 
 
-# wider than len df
-def test_inverse_transform_raises_value_error_when_wider_than_df() -> None:
-    wider_df = pd.DataFrame(
-        {
-            "variable_1": ["a", "b", "c"],
-            "variable_2": ["ZZ", "ZZ", "WW"],
-        }
-    )
-    coord, model = fit_transform(wider_df)
-    with pytest.raises(InvalidParameterException, match=r"n_dimensions"):
-        inverse_transform(coord, model)
-
-
 # using df with more dimensions than individuals and high column weights
 # allows for a more balanced probability in modality assignment during inverse transform
 
@@ -144,9 +130,7 @@ def test_inverse_transform_with_ponderation() -> None:
         "cont2": 1,
     }
     coord, model = fit_transform(df, col_weights=col_weights, seed=5)
-    result = inverse_transform(
-        coord, model, use_approximate_inverse=True, use_max_modalities=False
-    )
+    result = inverse_transform(coord, model, use_max_modalities=False)
     assert_frame_equal(result, inverse_expected)
 
 
@@ -167,9 +151,7 @@ def test_inverse_transform_deterministic() -> None:
         "cont2": 1,
     }
     coord, model = fit_transform(df, col_weights=col_weights)
-    result = inverse_transform(
-        coord, model, use_approximate_inverse=True, use_max_modalities=True
-    )
+    result = inverse_transform(coord, model, use_max_modalities=True)
     assert_frame_equal(result, inverse_expected)
 
 

saiph/projection_test.py

@@ -540,3 +540,30 @@ def test_transform_raise_error_on_wrong_columns(
     model = fit(df_to_fit)
     with pytest.raises(ColumnsNotFoundError):
         transform(df_to_transform, model)
+
+
+@pytest.mark.parametrize(
+    "df_to_fit_transform",
+    [
+        pd.DataFrame(
+            {"a": [1, 2, 3], "b": [4, 5, 6], "c": [7, 8, 9], "d": [10, 11, 12]}
+        ),  # pca
+        pd.DataFrame(
+            {"a": ["a", "b", "b"], "b": [1, 3, 6], "c": [1, 2, 3], "d": [1, 2, 3]}
+        ),  # famd
+        pd.DataFrame(
+            {
+                "a": ["a", "b", "b"],
+                "b": ["a", "b", "b"],
+                "c": ["a", "b", "b"],
+                "d": ["a", "b", "b"],
+            }
+        ),  # mca
+    ],
+)
+def test_fit_transform_with_horizontal_matrix(
+    df_to_fit_transform: pd.DataFrame,
+) -> None:
+    """Verify that the coordinates are a squared matrix even if the input is horizontal."""
+    coord, __ = fit_transform(df_to_fit_transform)
+    assert coord.shape[0] == coord.shape[1]

saiph/reduction/famd.py

@@ -188,7 +188,8 @@ def fit_transform(
             (more weight = more importance in the axes). default: np.ones(df.shape[1])
 
     Returns:
-        coord: The transformed data.
+        coord: The transformed data of size (n, min(n,p))
+        or (n, nf) if nf is specified.
         model: The model for transforming new data.
     """
     # If seed is None or int, we fit a Generator, else we use the one provided.
@@ -285,7 +286,8 @@ def transform(
         model: Model computed by fit.
 
     Returns:
-        coord: Coordinates of the dataframe in the fitted space.
+        coord: Coordinates of the dataframe in the fitted space of size (n, min(n,p))
+        or (n, nf) if nf is specified.
     """
     df_scaled = scaler(model, df)
     coord = pd.DataFrame(df_scaled @ model.V.T)

saiph/reduction/mca.py

@@ -133,7 +133,8 @@ def fit_transform(
 
     Returns:
         model: The model for transforming new data.
-        coord: The transformed data.
+        coord: The transformed data of size (n, min(n,p))
+        or (n, nf) if nf is specified.
     """
     random_gen = (
         seed if isinstance(seed, np.random.Generator) else np.random.default_rng(seed)
@@ -226,7 +227,8 @@ def transform(df: pd.DataFrame, model: Model) -> pd.DataFrame:
         model: Model computed by fit.
 
     Returns:
-        coord: Coordinates of the dataframe in the fitted space.
+        coord: Coordinates of the dataframe in the fitted space of size (n, min(n,p))
+        or (n, nf) if nf is specified.
     """
     df_scaled = scaler(model, df)
     coord = df_scaled @ model.D_c @ model.V.T

saiph/reduction/pca.py

@@ -100,7 +100,8 @@ def fit_transform(
 
     Returns:
         model: The model for transforming new data.
-        coord: The transformed data.
+        coord: The transformed data of size (n, min(n,p))
+        or (n, nf) if nf is specified.
     """
     model = fit(df, nf, col_weights, seed=seed)
     coord = transform(df, model)
@@ -171,7 +172,8 @@ def transform(df: pd.DataFrame, model: Model) -> pd.DataFrame:
         model: Model computed by fit.
 
     Returns:
-        coord: Coordinates of the dataframe in the fitted space.
+        coord: Coordinates of the dataframe in the fitted space of size (n, min(n,p))
+        or (n, nf) if nf is specified.
     """
     df_scaled = scaler(model, df)
     coord = df_scaled @ model.V.T