CLN Removed max_bins from splitter in GBDT (scikit-learn#13927)

Author: NicolasHug

sklearn/ensemble/_hist_gradient_boosting/grower.py

@@ -177,7 +177,7 @@ def __init__(self, X_binned, gradients, hessians, max_leaf_nodes=None,
  self.histogram_builder = HistogramBuilder(
  X_binned, max_bins, gradients, hessians, hessians_are_constant)
  self.splitter = Splitter(
- X_binned, max_bins, actual_n_bins, l2_regularization,
+ X_binned, actual_n_bins, l2_regularization,
  min_hessian_to_split, min_samples_leaf, min_gain_to_split,
  hessians_are_constant)
  self.max_leaf_nodes = max_leaf_nodes

sklearn/ensemble/_hist_gradient_boosting/splitting.pyx

@@ -91,9 +91,6 @@ cdef class Splitter:
  ----------
  X_binned : ndarray of int, shape (n_samples, n_features)
  The binned input samples. Must be Fortran-aligned.
- max_bins : int
- The maximum number of bins. Used to define the shape of the
- histograms.
  actual_n_bins : ndarray, shape (n_features,)
  The actual number of bins needed for each feature, which is lower or
  equal to max_bins.
@@ -114,7 +111,6 @@ cdef class Splitter:
  cdef public:
  const X_BINNED_DTYPE_C [::1, :] X_binned
  unsigned int n_features
- unsigned int max_bins
  unsigned int [::1] actual_n_bins
  unsigned char hessians_are_constant
  Y_DTYPE_C l2_regularization
@@ -126,18 +122,15 @@ cdef class Splitter:
  unsigned int [::1] left_indices_buffer
  unsigned int [::1] right_indices_buffer
 
- def __init__(self, const X_BINNED_DTYPE_C [::1, :] X_binned, unsigned int
- max_bins, np.ndarray[np.uint32_t] actual_n_bins,
+ def __init__(self, const X_BINNED_DTYPE_C [::1, :] X_binned,
+ np.ndarray[np.uint32_t] actual_n_bins,
  Y_DTYPE_C l2_regularization, Y_DTYPE_C
  min_hessian_to_split=1e-3, unsigned int
  min_samples_leaf=20, Y_DTYPE_C min_gain_to_split=0.,
  unsigned char hessians_are_constant=False):
 
  self.X_binned = X_binned
  self.n_features = X_binned.shape[1]
- # Note: all histograms will have <max_bins> bins, but some of the
- # last bins may be unused if actual_n_bins[f] < max_bins
- self.max_bins = max_bins
  self.actual_n_bins = actual_n_bins
  self.l2_regularization = l2_regularization
  self.min_hessian_to_split = min_hessian_to_split

sklearn/ensemble/_hist_gradient_boosting/tests/test_splitting.py

@@ -41,7 +41,6 @@ def test_histogram_split(n_bins):
  all_hessians,
  hessians_are_constant)
  splitter = Splitter(X_binned,
- n_bins,
  actual_n_bins,
  l2_regularization,
  min_hessian_to_split,
@@ -99,7 +98,7 @@ def test_gradient_and_hessian_sanity(constant_hessian):
  dtype=np.uint32)
  builder = HistogramBuilder(X_binned, n_bins, all_gradients,
  all_hessians, constant_hessian)
- splitter = Splitter(X_binned, n_bins, actual_n_bins,
+ splitter = Splitter(X_binned, actual_n_bins,
  l2_regularization, min_hessian_to_split,
  min_samples_leaf, min_gain_to_split, constant_hessian)
 
@@ -196,7 +195,7 @@ def test_split_indices():
  builder = HistogramBuilder(X_binned, n_bins,
  all_gradients, all_hessians,
  hessians_are_constant)
- splitter = Splitter(X_binned, n_bins, actual_n_bins,
+ splitter = Splitter(X_binned, actual_n_bins,
  l2_regularization, min_hessian_to_split,
  min_samples_leaf, min_gain_to_split,
  hessians_are_constant)
@@ -251,7 +250,7 @@ def test_min_gain_to_split():
  dtype=np.uint32)
  builder = HistogramBuilder(X_binned, n_bins, all_gradients,
  all_hessians, hessians_are_constant)
- splitter = Splitter(X_binned, n_bins, actual_n_bins,
+ splitter = Splitter(X_binned, actual_n_bins,
  l2_regularization, min_hessian_to_split,
  min_samples_leaf, min_gain_to_split,
  hessians_are_constant)