fix types multidim scaling (#646)

* fix types multidim scaling

* fix types

* add error type

Author: arnaudvl

alibi/explainers/cfproto.py

@@ -38,7 +38,7 @@ def __init__(self,
                  shape: tuple,
                  kappa: float = 0.,
                  beta: float = .1,
-                 feature_range: tuple = (-1e10, 1e10),
+                 feature_range: Tuple[Union[float, np.ndarray], Union[float, np.ndarray]] = (-1e10, 1e10),
                  gamma: float = 0.,
                  ae_model: Optional[tf.keras.Model] = None,
                  enc_model: Optional[tf.keras.Model] = None,
@@ -178,7 +178,9 @@ def __init__(self,
         self.max_iterations = max_iterations
         self.c_init = c_init
         self.c_steps = c_steps
-        self.feature_range = feature_range
+        self.feature_range = tuple([(np.ones(shape[1:]) * feature_range[_])[None, :]
+                                    if isinstance(feature_range[_], float) else feature_range[_]
+                                    for _ in range(2)])
         self.update_num_grad = update_num_grad
         self.eps = eps
         self.clip = clip
@@ -754,13 +756,13 @@ def fit(self,
 
                 # multidim scaled distances
                 d_abs_abdm, _ = multidim_scaling(d_abdm, n_components=2, use_metric=True,
-                                                 feature_range=self.feature_range,
+                                                 feature_range=self.feature_range,  # type: ignore[arg-type]
                                                  standardize_cat_vars=standardize_cat_vars,
                                                  smooth=smooth, center=center,
                                                  update_feature_range=False)
 
                 d_abs_mvdm, _ = multidim_scaling(d_mvdm, n_components=2, use_metric=True,
-                                                 feature_range=self.feature_range,
+                                                 feature_range=self.feature_range,  # type: ignore[arg-type]
                                                  standardize_cat_vars=standardize_cat_vars,
                                                  smooth=smooth, center=center,
                                                  update_feature_range=False)
@@ -779,7 +781,7 @@ def fit(self,
                     self.feature_range = new_feature_range
             else:  # apply multidimensional scaling for the abdm or mvdm distances
                 self.d_abs, self.feature_range = multidim_scaling(d_pair, n_components=2, use_metric=True,
-                                                                  feature_range=self.feature_range,
+                                                                  feature_range=self.feature_range,  # type: ignore
                                                                   standardize_cat_vars=standardize_cat_vars,
                                                                   smooth=smooth, center=center,
                                                                   update_feature_range=update_feature_range)

alibi/utils/distance.py

@@ -162,7 +162,7 @@ def abdm(X: np.ndarray,
 
 
 def multidim_scaling(d_pair: dict,
-                     feature_range: tuple,
+                     feature_range: Tuple[np.ndarray, np.ndarray],
                      n_components: int = 2,
                      use_metric: bool = True,
                      standardize_cat_vars: bool = True,
@@ -178,8 +178,8 @@ def multidim_scaling(d_pair: dict,
         Dict with as keys the column index of the categorical variables and as values
         a pairwise distance matrix for the categories of the variable.
     feature_range
-        Tuple with `min` and `max` ranges to allow for perturbed instances. `Min` and `max` ranges can be `float` or
-        `numpy` arrays with dimension (`1 x nb of features`) for feature-wise ranges.
+        Tuple with `min` and `max` ranges to allow for perturbed instances. `Min` and `max` ranges are
+        `numpy` arrays with dimension (`1 x nb of features`).
     n_components
         Number of dimensions in which to immerse the dissimilarities.
     use_metric
@@ -240,6 +240,6 @@ def multidim_scaling(d_pair: dict,
         d_abs_scaled[k] = d_scaled  # scaled distance from the origin for each category
 
     if update_feature_range:
-        feature_range = new_feature_range
+        feature_range = new_feature_range  # type: ignore
 
     return d_abs_scaled, feature_range