MAINT Submoduletest

sktree/tree/_marginal.pxd

@@ -3,8 +3,7 @@ import numpy as np
 cimport numpy as cnp
 
 from .._lib.sklearn.tree._tree cimport BaseTree, Node
-from .._lib.sklearn.tree._utils cimport UINT32_t
-from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, intp_t
+from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, intp_t, uint32_t
 
 
 cpdef apply_marginal_tree(

sktree/tree/_marginal.pyx

@@ -65,7 +65,7 @@ cpdef apply_marginal_tree(
     cdef intp_t n_marginals = marginal_indices.shape[0]
 
     # sklearn_rand_r random number state
-    cdef UINT32_t rand_r_state = random_state.randint(0, RAND_R_MAX)
+    cdef uint32_t rand_r_state = random_state.randint(0, RAND_R_MAX)
 
     # define a set of all marginal indices
     cdef unordered_set[intp_t] marginal_indices_map
@@ -108,7 +108,7 @@ cdef inline cnp.ndarray _apply_dense_marginal(
     unordered_set[intp_t] marginal_indices_map,
     intp_t traversal_method,
     unsigned char use_sample_weight,
-    UINT32_t* rand_r_state
+    uint32_t* rand_r_state
 ):
     """Finds the terminal region (=leaf node) for each sample in X.
 
@@ -131,7 +131,7 @@ cdef inline cnp.ndarray _apply_dense_marginal(
     use_sample_weight : unsigned char
         Whether or not to use the weighted number of samples
         in each node.
-    rand_r_state : UINT32_t
+    rand_r_state : uint32_t
         The random number state.
     """
     # Extract input

sktree/tree/_oblique_splitter.pxd

@@ -10,13 +10,12 @@
 
 import numpy as np
 
-cimport numpy as cnp
 from libcpp.vector cimport vector
 
 from .._lib.sklearn.tree._criterion cimport Criterion
 from .._lib.sklearn.tree._splitter cimport SplitRecord, Splitter
-from .._lib.sklearn.tree._utils cimport UINT32_t
-from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, intp_t
+from .._lib.sklearn.tree._tree cimport ParentInfo
+from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, int8_t, intp_t, uint32_t
 from ._sklearn_splitter cimport sort
 
 
@@ -30,11 +29,8 @@ cdef struct ObliqueSplitRecord:
     float64_t improvement          # Impurity improvement given parent node.
     float64_t impurity_left        # Impurity of the left split.
     float64_t impurity_right       # Impurity of the right split.
-    float64_t lower_bound       # Lower bound on value of both children for monotonicity
-    float64_t upper_bound       # Upper bound on value of both children for monotonicity
     unsigned char missing_go_to_left  # Controls if missing values go to the left node.
     intp_t n_missing            # Number of missing values for the feature being split on
-    intp_t n_constant_features  # Number of constant features in the split
 
     vector[float32_t]* proj_vec_weights  # weights of the vector (max_features,)
     vector[intp_t]* proj_vec_indices     # indices of the features (max_features,)
@@ -83,17 +79,15 @@ cdef class BaseObliqueSplitter(Splitter):
 
     cdef int node_split(
         self,
-        float64_t impurity,   # Impurity of the node
+        ParentInfo* parent,
         SplitRecord* split,
-        float64_t lower_bound,
-        float64_t upper_bound,
     ) except -1 nogil
 
     cdef inline void fisher_yates_shuffle_memview(
         self,
         intp_t[::1] indices_to_sample,
         intp_t grid_size,
-        UINT32_t* random_state
+        uint32_t* random_state
     ) noexcept nogil
 
 cdef class ObliqueSplitter(BaseObliqueSplitter):
@@ -118,10 +112,8 @@ cdef class ObliqueSplitter(BaseObliqueSplitter):
 cdef class BestObliqueSplitter(ObliqueSplitter):
     cdef int node_split(
         self,
-        float64_t impurity,   # Impurity of the node
+        ParentInfo* parent,
         SplitRecord* split,
-        float64_t lower_bound,
-        float64_t upper_bound,
     ) except -1 nogil
 
 
@@ -140,10 +132,8 @@ cdef class RandomObliqueSplitter(ObliqueSplitter):
 
     cdef int node_split(
         self,
-        float64_t impurity,   # Impurity of the node
+        ParentInfo* parent,
         SplitRecord* split,
-        float64_t lower_bound,
-        float64_t upper_bound,
     ) except -1 nogil
 
 

sktree/tree/_oblique_splitter.pyx

@@ -30,6 +30,8 @@ cdef inline void _init_split(ObliqueSplitRecord* self, intp_t start_pos) noexcep
     self.feature = 0
     self.threshold = 0.
     self.improvement = -INFINITY
+    self.missing_go_to_left = False
+    self.n_missing = 0
 
 
 cdef class BaseObliqueSplitter(Splitter):
@@ -128,7 +130,7 @@ cdef class BaseObliqueSplitter(Splitter):
         self,
         intp_t[::1] indices_to_sample,
         intp_t grid_size,
-        UINT32_t* random_state,
+        uint32_t* random_state,
     ) noexcept nogil:
         cdef intp_t i, j
 
@@ -146,7 +148,7 @@ cdef class ObliqueSplitter(BaseObliqueSplitter):
         intp_t min_samples_leaf,
         float64_t min_weight_leaf,
         object random_state,
-        const cnp.int8_t[:] monotonic_cst,
+        const int8_t[:] monotonic_cst,
         float64_t feature_combinations,
         *argv
     ):
@@ -245,7 +247,7 @@ cdef class ObliqueSplitter(BaseObliqueSplitter):
         """
         cdef intp_t n_features = self.n_features
         cdef intp_t n_non_zeros = self.n_non_zeros
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
 
         cdef intp_t i, feat_i, proj_i, rand_vec_index
         cdef float32_t weight
@@ -290,10 +292,8 @@ cdef class BestObliqueSplitter(ObliqueSplitter):
 
     cdef int node_split(
         self,
-        float64_t impurity,
+        ParentInfo* parent_record,
         SplitRecord* split,
-        float64_t lower_bound,
-        float64_t upper_bound,
     ) except -1 nogil:
         """Find the best_split split on node samples[start:end]
 
@@ -319,6 +319,11 @@ cdef class BestObliqueSplitter(ObliqueSplitter):
         cdef float64_t current_proxy_improvement = -INFINITY
         cdef float64_t best_proxy_improvement = -INFINITY
 
+        with gil:
+            print("Accessing parent record")
+
+        cdef float64_t impurity = parent_record.impurity
+
         cdef intp_t feat_i, p       # index over computed features and start/end
         cdef intp_t partition_end
         cdef float32_t temp_d         # to compute a projection feature value
@@ -430,7 +435,14 @@ cdef class BestObliqueSplitter(ObliqueSplitter):
         deref(oblique_split).improvement = best_split.improvement
         deref(oblique_split).impurity_left = best_split.impurity_left
         deref(oblique_split).impurity_right = best_split.impurity_right
-        deref(oblique_split).n_constant_features = 0
+
+        with gil:
+            print("Segfaulting here...")
+        # XXX: Fix
+        parent_record.n_constant_features = 0
+
+        with gil:
+            print("Got past")
         return 0
 
 cdef class RandomObliqueSplitter(ObliqueSplitter):
@@ -496,10 +508,8 @@ cdef class RandomObliqueSplitter(ObliqueSplitter):
     # overwrite the node_split method with random threshold selection
     cdef int node_split(
         self,
-        float64_t impurity,
+        ParentInfo* parent_record,
         SplitRecord* split,
-        float64_t lower_bound,
-        float64_t upper_bound,
     ) except -1 nogil:
         """Find the best_split split on node samples[start:end]
 
@@ -513,7 +523,7 @@ cdef class RandomObliqueSplitter(ObliqueSplitter):
         cdef intp_t[::1] samples = self.samples
         cdef intp_t start = self.start
         cdef intp_t end = self.end
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
 
         # pointer array to store feature values to split on
         cdef float32_t[::1] feature_values = self.feature_values
@@ -534,6 +544,8 @@ cdef class RandomObliqueSplitter(ObliqueSplitter):
         cdef float32_t min_feature_value
         cdef float32_t max_feature_value
 
+        cdef float64_t impurity = parent_record.impurity
+
         # Number of features discovered to be constant during the split search
         # cdef intp_t n_found_constants = 0
         # cdef intp_t n_known_constants = n_constant_features[0]
@@ -655,7 +667,9 @@ cdef class RandomObliqueSplitter(ObliqueSplitter):
         deref(oblique_split).improvement = best_split.improvement
         deref(oblique_split).impurity_left = best_split.impurity_left
         deref(oblique_split).impurity_right = best_split.impurity_right
-        # deref(oblique_split).n_constant_features = 0
+
+        # XXX: Fix
+        parent_record.n_constant_features = 0
         return 0
 
 
@@ -667,7 +681,7 @@ cdef class MultiViewSplitter(BestObliqueSplitter):
         intp_t min_samples_leaf,
         float64_t min_weight_leaf,
         object random_state,
-        const cnp.int8_t[:] monotonic_cst,
+        const int8_t[:] monotonic_cst,
         float64_t feature_combinations,
         const intp_t[:] feature_set_ends,
         intp_t n_feature_sets,
@@ -740,7 +754,7 @@ cdef class MultiViewSplitter(BestObliqueSplitter):
         This proceeds as a normal sampling projection matrix,
         but now also uniformly samples features from each feature set.
         """
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
         cdef intp_t feat_i, proj_i
         cdef float32_t weight
 
@@ -839,7 +853,7 @@ cdef class MultiViewObliqueSplitter(BestObliqueSplitter):
         intp_t min_samples_leaf,
         float64_t min_weight_leaf,
         object random_state,
-        const cnp.int8_t[:] monotonic_cst,
+        const int8_t[:] monotonic_cst,
         float64_t feature_combinations,
         const intp_t[:] feature_set_ends,
         intp_t n_feature_sets,
@@ -914,7 +928,7 @@ cdef class MultiViewObliqueSplitter(BestObliqueSplitter):
         """
         cdef intp_t n_features = self.n_features
         cdef intp_t n_non_zeros = self.n_non_zeros
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
 
         cdef intp_t i, j, feat_i, proj_i, rand_vec_index
         cdef float32_t weight

sktree/tree/_utils.pxd

@@ -5,11 +5,10 @@ cimport numpy as cnp
 cnp.import_array()
 
 from .._lib.sklearn.tree._splitter cimport SplitRecord
-from .._lib.sklearn.tree._utils cimport UINT32_t
-from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, int32_t, intp_t
+from .._lib.sklearn.utils._typedefs cimport float32_t, float64_t, int32_t, intp_t, uint32_t
 
 
-cdef int rand_weighted_binary(float64_t p0, UINT32_t* random_state) noexcept nogil
+cdef int rand_weighted_binary(float64_t p0, uint32_t* random_state) noexcept nogil
 
 cpdef unravel_index(
     intp_t index, cnp.ndarray[intp_t, ndim=1] shape

sktree/tree/_utils.pyx

@@ -14,14 +14,14 @@ cnp.import_array()
 from .._lib.sklearn.tree._utils cimport rand_uniform
 
 
-cdef inline int rand_weighted_binary(float64_t p0, UINT32_t* random_state) noexcept nogil:
+cdef inline int rand_weighted_binary(float64_t p0, uint32_t* random_state) noexcept nogil:
     """Sample from integers 0 and 1 with different probabilities.
 
     Parameters
     ----------
     p0 : float64_t
         The probability of sampling 0.
-    random_state : UINT32_t*
+    random_state : uint32_t*
         The random state.
     """
     cdef float64_t random_value = rand_uniform(0.0, 1.0, random_state)

sktree/tree/manifold/_morf_splitter.pxd

@@ -10,12 +10,10 @@
 
 import numpy as np
 
-cimport numpy as cnp
 from libcpp.vector cimport vector
 
 from ..._lib.sklearn.tree._splitter cimport SplitRecord
-from ..._lib.sklearn.tree._utils cimport UINT32_t
-from ..._lib.sklearn.utils._typedefs cimport float32_t, float64_t, intp_t
+from ..._lib.sklearn.utils._typedefs cimport float32_t, float64_t, int8_t, intp_t, uint8_t, uint32_t
 from .._oblique_splitter cimport BestObliqueSplitter, ObliqueSplitRecord
 
 # https://github.com/cython/cython/blob/master/Cython/Includes/libcpp/algorithm.pxd
@@ -29,12 +27,6 @@ from .._oblique_splitter cimport BestObliqueSplitter, ObliqueSplitRecord
 #         discrete_distribution(T first, T last) except +
 #         operator()(&G) except +
 
-# XXX: replace with from libcpp.algorithm cimport swap
-# when Cython 3.0 is released
-cdef extern from "<algorithm>" namespace "std" nogil:
-    void swap[T](T& a, T& b) except +  # array overload also works
-
-
 cdef class PatchSplitter(BestObliqueSplitter):
     # The PatchSplitter creates candidate feature values by sampling 2D patches from
     # an input data vector. The input data is vectorized, so `data_height` and
@@ -53,7 +45,7 @@ cdef class PatchSplitter(BestObliqueSplitter):
     cdef const intp_t[:] data_dims                      # The dimensions of the input data
     cdef const intp_t[:] min_patch_dims                 # The minimum size of the patch to sample in each dimension
     cdef const intp_t[:] max_patch_dims                 # The maximum size of the patch to sample in each dimension
-    cdef const cnp.uint8_t[:] dim_contiguous            # A boolean array indicating whether each dimension is contiguous
+    cdef const uint8_t[:] dim_contiguous            # A boolean array indicating whether each dimension is contiguous
 
     # TODO: check if this works and is necessary for discontiguous data
     # cdef intp_t[:] stride_offsets                # The stride offsets for each dimension

sktree/tree/manifold/_morf_splitter.pyx

@@ -7,10 +7,6 @@
 
 import numpy as np
 
-cimport numpy as cnp
-
-cnp.import_array()
-
 from cython.operator cimport dereference as deref
 from libcpp.vector cimport vector
 
@@ -124,11 +120,11 @@ cdef class BestPatchSplitter(BaseDensePatchSplitter):
         intp_t min_samples_leaf,
         float64_t min_weight_leaf,
         object random_state,
-        const cnp.int8_t[:] monotonic_cst,
+        const int8_t[:] monotonic_cst,
         float64_t feature_combinations,
         const intp_t[:] min_patch_dims,
         const intp_t[:] max_patch_dims,
-        const cnp.uint8_t[:] dim_contiguous,
+        const uint8_t[:] dim_contiguous,
         const intp_t[:] data_dims,
         bytes boundary,
         const float32_t[:, :] feature_weight,
@@ -212,7 +208,7 @@ cdef class BestPatchSplitter(BaseDensePatchSplitter):
         cdef intp_t top_left_patch_seed
         cdef intp_t patch_size = 1
 
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
 
         # define parameters for the random patch
         cdef intp_t patch_dim
@@ -315,7 +311,7 @@ cdef class BestPatchSplitter(BaseDensePatchSplitter):
         intp_t top_left_patch_seed,
         const intp_t[:] patch_dims,
     ) noexcept nogil:
-        cdef UINT32_t* random_state = &self.rand_r_state
+        cdef uint32_t* random_state = &self.rand_r_state
         # iterates over the size of the patch
         cdef intp_t patch_idx