[sml] optimize preprocessing by eliminating unnecessary where function (

#622)

# Pull Request

## What problem does this PR solve?
Small optimization in sml/preprocessing
Eliminate `where` function in original solution.
In original version, `where` was used only for eliminating redundant
computation branch, however both branches within the `where` function
still need to be computed, which does not meet the desired outcome.
By eliminating the `where` function, the computation can be reduced
without affecting the result.

sml/preprocessing/preprocessing.py

@@ -838,7 +838,7 @@ def bin_func(x, KMEANS):
         elif vectorize == True:
             diverse_n_bins = self.diverse_n_bins
             ### directly using jnp.linspace will cause dynamic shape problem,
-            ### so we need to use jnp.arange and a branch function jnp.where
+            ### so we need to use jnp.arange with a public value n_bins
             if self.strategy == "uniform":
                 arrange_array = jnp.arange(n_bins + 1)
 
@@ -848,9 +848,7 @@ def bin_func(x, diverse_n_bin, arrange_array):
                     delta = (maxval - minval) / diverse_n_bin
 
                     def bin_element_func(x_inner):
-                        return jnp.where(
-                            x_inner <= diverse_n_bin, minval + x_inner * delta, maxval
-                        )
+                        return minval + x_inner * delta
 
                     return jax.vmap(bin_element_func)(arrange_array)