readme results made coherent

Author: Julien Roussel

README.rst

@@ -88,9 +88,8 @@ With just these few lines of code, you can see how easy it is to
   generator_holes = missing_patterns.EmpiricalHoleGenerator(n_splits=4, ratio_masked=0.1)
   comparison = comparator.Comparator(
         dict_imputers,
-        columns,
         generator_holes = generator_holes,
-        metrics = ["mae", "wmape", "kl_columnwise", "ks_test", "energy"],
+        metrics = ["mae", "wmape", "kl_columnwise", "frechet"],
     )
   results = comparison.compare(df_with_nan)
   results.style.highlight_min(color="lightsteelblue", axis=1)

examples/tutorials/plot_tuto_diffusion_models.py

@@ -169,7 +169,6 @@
 
 comparison = comparator.Comparator(
     dict_imputers,
-    selected_columns=df_data.columns,
     generator_holes=missing_patterns.UniformHoleGenerator(n_splits=2, random_state=rng),
     metrics=["mae", "kl_columnwise"],
 )
@@ -224,7 +223,6 @@
 
 comparison = comparator.Comparator(
     dict_imputers,
-    selected_columns=df_data.columns,
     generator_holes=missing_patterns.UniformHoleGenerator(n_splits=2, random_state=rng),
     metrics=["mae", "kl_columnwise"],
 )

qolmat/benchmark/comparator.py

@@ -28,9 +28,6 @@ class Comparator:
     ----------
     dict_models: Dict[str, any]
         dictionary of imputation methods
-    selected_columns: List[str]Œ
-        list of column's names selected (all with at least one null value will
-        be imputed)
     columnwise_evaluation : Optional[bool], optional
         whether the metric should be calculated column-wise or not,
         by default False
@@ -46,7 +43,6 @@ class Comparator:
     def __init__(
         self,
         dict_models: Dict[str, Any],
-        selected_columns: List[str],
         generator_holes: _HoleGenerator,
         metrics: List = ["mae", "wmape", "kl_columnwise"],
         dict_config_opti: Optional[Dict[str, Any]] = {},
@@ -55,7 +51,6 @@ def __init__(
         verbose: bool = False,
     ):
         self.dict_imputers = dict_models
-        self.selected_columns = selected_columns
         self.generator_holes = generator_holes
         self.metrics = metrics
         self.dict_config_opti = dict_config_opti

qolmat/benchmark/metrics.py

@@ -835,6 +835,7 @@ def sum_pairwise_distances(
 def frechet_distance_base(
     df1: pd.DataFrame,
     df2: pd.DataFrame,
+    df_mask: pd.DataFrame,
 ) -> pd.Series:
     """Compute the Fréchet distance between two dataframes df1 and df2.
 
@@ -853,16 +854,24 @@ def frechet_distance_base(
         true dataframe
     df2 : pd.DataFrame
         predicted dataframe
+    df_mask : pd.DataFrame
+        Elements of the dataframes to compute on
 
     Returns
     -------
     pd.Series
         Frechet distance in a Series object
 
     """
-    if df1.shape != df2.shape:
+    if df1.shape != df2.shape or df1.shape != df_mask.shape:
         raise Exception("inputs have to be of same dimensions.")
 
+    df1 = df1.copy()
+    df2 = df2.copy()
+    # Set to nan the values not in the mask
+    df1[~df_mask] = np.nan
+    df2[~df_mask] = np.nan
+
     std = (np.std(df1) + np.std(df2) + EPS) / 2
     mu = (np.nanmean(df1, axis=0) + np.nanmean(df2, axis=0)) / 2
     df1 = (df1 - mu) / std
@@ -911,7 +920,7 @@ def frechet_distance(
 
     """
     if method == "single":
-        return frechet_distance_base(df1, df2)
+        return frechet_distance_base(df1, df2, df_mask)
     return pattern_based_weighted_mean_metric(
         df1,
         df2,

qolmat/imputations/imputers_pytorch.py

@@ -8,6 +8,7 @@
 import pandas as pd
 from numpy.typing import NDArray
 from sklearn.preprocessing import StandardScaler
+from tqdm import tqdm
 
 # from typing_extensions import Self
 from qolmat.benchmark import metrics
@@ -106,23 +107,29 @@ def _fit_estimator(
         optimizer = optim.Adam(estimator.parameters(), lr=self.learning_rate)
         loss_fn = self.loss_fn
 
-        for epoch in range(self.epochs):
-            estimator.train()
-            optimizer.zero_grad()
-
-            input_data = torch.Tensor(X.values)
-            target_data = torch.Tensor(y.values)
-            target_data = target_data.unsqueeze(1)
-            outputs = estimator(input_data)
-            loss = loss_fn(outputs, target_data)
-
-            loss.backward()
-            optimizer.step()
-            if (epoch + 1) % 10 == 0:
-                logging.info(
-                    f"Epoch [{epoch + 1}/{self.epochs}], "
-                    f"Loss: {loss.item():.4f}"
-                )
+        # if X.shape[0] != estimator[0].in_features:
+        #     raise ValueError(
+        #         "The number of features in X does not match the input "
+        #         "features of the estimator. The estimator expects"
+        #         f" {estimator[0].in_features} features, but X has "
+        #         f"{X.shape[0]} features."
+        #     )
+
+        with tqdm(total=self.epochs, desc="Training", unit="epoch") as pbar:
+            for _ in range(self.epochs):
+                estimator.train()
+                optimizer.zero_grad()
+
+                input_data = torch.Tensor(X.values)
+                target_data = torch.Tensor(y.values)
+                target_data = target_data.unsqueeze(1)
+                outputs = estimator(input_data)
+                loss = loss_fn(outputs, target_data)
+
+                loss.backward()
+                optimizer.step()
+                pbar.set_postfix(loss=f"{loss.item():.4f}")
+                pbar.update(1)
         return estimator
 
     def _predict_estimator(

qolmat/imputations/rpca/rpca_noisy.py

@@ -317,67 +317,73 @@ def minimise_loss(
         Ir = np.eye(rank)
         In = identity(n_rows)
 
-        for _ in tqdm(
-            range(max_iterations),
+        with tqdm(
+            total=max_iterations,
             desc="Noisy RPCA loss minimization",
+            unit="iteration",
             disable=not verbose,
-        ):
-            M_temp = M.copy()
-            A_temp = A.copy()
-            L_temp = L.copy()
-            Q_temp = Q.copy()
-            if norm == "L1":
-                R_temp = R.copy()
-                sums = np.zeros((n_rows, n_cols))
-                for i_period, _ in enumerate(list_periods):
-                    sums += mu * R[i_period] - list_H[i_period] @ Y
-
-                M = spsolve(
-                    (1 + mu) * In + HtH,
-                    D - A + mu * L @ Q - Y + sums,
-                )
-            else:
-                M = spsolve(
-                    (1 + mu) * In + 2 * HtH,
-                    D - A + mu * L @ Q - Y,
-                )
-            M = M.reshape(D.shape)
-
-            A_Omega = rpca_utils.soft_thresholding(D - M, lam)
-            A_Omega_C = D - M
-            A = np.where(Omega, A_Omega, A_Omega_C)
-            Q = scp.linalg.solve(
-                a=tau * Ir + mu * (L.T @ L),
-                b=L.T @ (mu * M + Y),
-            )
-
-            L = scp.linalg.solve(
-                a=tau * Ir + mu * (Q @ Q.T),
-                b=Q @ (mu * M.T + Y.T),
-            ).T
-
-            Y += mu * (M - L @ Q)
-            if norm == "L1":
-                for i_period, _ in enumerate(list_periods):
-                    eta = list_etas[i_period]
-                    R[i_period] = rpca_utils.soft_thresholding(
-                        R[i_period] / mu, eta / mu
+        ) as pbar:
+            for _ in range(max_iterations):
+                M_temp = M.copy()
+                A_temp = A.copy()
+                L_temp = L.copy()
+                Q_temp = Q.copy()
+                if norm == "L1":
+                    R_temp = R.copy()
+                    sums = np.zeros((n_rows, n_cols))
+                    for i_period, _ in enumerate(list_periods):
+                        sums += mu * R[i_period] - list_H[i_period] @ Y
+
+                    M = spsolve(
+                        (1 + mu) * In + HtH,
+                        D - A + mu * L @ Q - Y + sums,
                     )
+                else:
+                    M = spsolve(
+                        (1 + mu) * In + 2 * HtH,
+                        D - A + mu * L @ Q - Y,
+                    )
+                M = M.reshape(D.shape)
+
+                A_Omega = rpca_utils.soft_thresholding(D - M, lam)
+                A_Omega_C = D - M
+                A = np.where(Omega, A_Omega, A_Omega_C)
+                Q = scp.linalg.solve(
+                    a=tau * Ir + mu * (L.T @ L),
+                    b=L.T @ (mu * M + Y),
+                )
 
-            mu = min(mu * rho, mu_bar)
-
-            Mc = np.linalg.norm(M - M_temp, np.inf)
-            Ac = np.linalg.norm(A - A_temp, np.inf)
-            Lc = np.linalg.norm(L - L_temp, np.inf)
-            Qc = np.linalg.norm(Q - Q_temp, np.inf)
-            error_max = max([Mc, Ac, Lc, Qc])  # type: ignore # noqa
-            if norm == "L1":
-                for i_period, _ in enumerate(list_periods):
-                    Rc = np.linalg.norm(R[i_period] - R_temp[i_period], np.inf)
-                    error_max = max(error_max, Rc)  # type: ignore # noqa
-
-            if error_max < tolerance:
-                break
+                L = scp.linalg.solve(
+                    a=tau * Ir + mu * (Q @ Q.T),
+                    b=Q @ (mu * M.T + Y.T),
+                ).T
+
+                Y += mu * (M - L @ Q)
+                if norm == "L1":
+                    for i_period, _ in enumerate(list_periods):
+                        eta = list_etas[i_period]
+                        R[i_period] = rpca_utils.soft_thresholding(
+                            R[i_period] / mu, eta / mu
+                        )
+
+                mu = min(mu * rho, mu_bar)
+
+                Mc = np.linalg.norm(M - M_temp, np.inf)
+                Ac = np.linalg.norm(A - A_temp, np.inf)
+                Lc = np.linalg.norm(L - L_temp, np.inf)
+                Qc = np.linalg.norm(Q - Q_temp, np.inf)
+                error_max = max([Mc, Ac, Lc, Qc])  # type: ignore # noqa
+                if norm == "L1":
+                    for i_period, _ in enumerate(list_periods):
+                        Rc = np.linalg.norm(
+                            R[i_period] - R_temp[i_period], np.inf
+                        )
+                        error_max = max(error_max, Rc)  # type: ignore # noqa
+
+                if error_max < tolerance:
+                    break
+                pbar.set_postfix(error=f"{error_max.item():.4f}")
+                pbar.update(1)
 
         M = L @ Q
 

tests/benchmark/test_comparator.py

@@ -40,7 +40,6 @@ def generator_holes_mock(mocker: MockerFixture) -> _HoleGenerator:
 def comparator(generator_holes_mock: _HoleGenerator) -> Comparator:
     return Comparator(
         dict_models={},
-        selected_columns=["A", "B"],
         generator_holes=generator_holes_mock,
         metrics=["mae", "mse"],
     )
@@ -439,7 +438,6 @@ def test_compare_reproducibility():
     )
     comparator = Comparator(
         dict_models=dict_models,
-        selected_columns=df_data.columns,
         generator_holes=generator_holes,
         metrics=["mae", "mse"],
     )

tests/benchmark/test_metrics.py

@@ -171,12 +171,16 @@ def test_kl_divergence_gaussian(
 
 @pytest.mark.parametrize("df1", [df_incomplete])
 @pytest.mark.parametrize("df2", [df_imputed])
-def test_frechet_distance_base(df1: pd.DataFrame, df2: pd.DataFrame) -> None:
-    result = metrics.frechet_distance_base(df1, df1)
+@pytest.mark.parametrize("df_mask", [df_mask])
+def test_frechet_distance_base(
+    df1: pd.DataFrame, df2: pd.DataFrame, df_mask: pd.DataFrame
+) -> None:
+    result = metrics.frechet_distance_base(df1, df1, df_mask)
     np.testing.assert_allclose(result, 0, atol=1e-3)
 
-    result = metrics.frechet_distance_base(df1, df2)
-    np.testing.assert_allclose(result, 0.134, atol=1e-3)
+    result = metrics.frechet_distance_base(df1, df2, df_mask)
+    assert np.all(0 < result)
+    assert np.all(result < 1)
 
 
 @pytest.mark.parametrize("df1", [df_incomplete])