Sklearn decisiontree (#23565)

Co-authored-by: umairjavaid <[email protected]>

ivy/functional/frontends/sklearn/tree/_tree.py

@@ -1,31 +1,33 @@
 import ivy
-
-
 import numpy as np
-
-
 from scipy.sparse import issparse
 from scipy.sparse import csr_matrix
 from scipy.sparse import isspmatrix_csr
 
 
+
 # =============================================================================
 # Types and constants
 # =============================================================================
 
 from numpy import float32 as DTYPE
 from numpy import float64 as DOUBLE
 
-EPSILON = np.finfo(np.double).eps
+
 # Define constants
 INFINITY = np.inf
+EPSILON = np.finfo(np.double).eps
+
 # Some handy constants (BestFirstTreeBuilder)
 IS_FIRST = 1
-IS_LEFT = 1
 IS_NOT_FIRST = 0
+IS_LEFT = 1
 IS_NOT_LEFT = 0
+
 TREE_LEAF = -1
 TREE_UNDEFINED = -2
+_TREE_LEAF = TREE_LEAF
+_TREE_UNDEFINED = TREE_UNDEFINED
 
 
 class SplitRecord:
@@ -49,14 +51,17 @@ def __init__(self):
         self.missing_go_to_left = None
 
 
+dummy = Node()
+# Create a numpy dtype for Node using the dummy object
+NODE_DTYPE = np.asarray([dummy], dtype=object).dtype
+
 # =============================================================================
 # TreeBuilder
 # =============================================================================
 
 
 class TreeBuilder:
     """Interface for different tree building strategies."""
-
     def __init__(self):
         self.splitter = None
         self.min_samples_split = None
@@ -82,12 +87,10 @@ def _check_input(
         y,
         sample_weight,
     ):
-        """Check input dtype, layout, and format."""
+        """Check input dtype, layout, and format"""
         if issparse(X):
-            X = (
-                X.tocsc()
-            )  # tocsc() is a method provided by the scipy.sparse module in the SciPy library. It's used to convert a sparse matrix to the Compressed Sparse Column (CSC) format.
-            X.sort_indices()  # This is done to ensure that the indices of non-zero elements within the matrix are sorted in ascending order.
+            X = X.tocsc() #tocsc() is a method provided by the scipy.sparse module in the SciPy library. It's used to convert a sparse matrix to the Compressed Sparse Column (CSC) format. 
+            X.sort_indices() #This is done to ensure that the indices of non-zero elements within the matrix are sorted in ascending order.
 
             if X.data.dtype != DTYPE:
                 X.data = np.ascontiguousarray(X.data, dtype=DTYPE)
@@ -126,6 +129,144 @@ def __init__(
         self.n_constant_features = n_constant_features
 
 
+class DepthFirstTreeBuilder(TreeBuilder):
+    """Build a decision tree in depth-first fashion."""
+
+    def __init__(
+        self, splitter, min_samples_split,
+        min_samples_leaf, min_weight_leaf,
+        max_depth, min_impurity_decrease
+    ):
+        self.splitter = splitter
+        self.min_samples_split = min_samples_split
+        self.min_samples_leaf = min_samples_leaf
+        self.min_weight_leaf = min_weight_leaf
+        self.max_depth = max_depth
+        self.min_impurity_decrease = min_impurity_decrease
+
+    def build(
+        self,
+        tree,
+        X,
+        y,
+        sample_weight=None,
+        missing_values_in_feature_mask=None
+    ):
+        """Build a decision tree from the training set (X, y)."""
+
+        # Check input
+        X, y, sample_weight = self._check_input(X, y, sample_weight)
+
+        # Initial capacity
+        init_capacity = (2 ** (tree.max_depth + 1)) - 1 if tree.max_depth <= 10 else 2047
+
+        tree._resize(init_capacity)
+
+        # Parameters
+        splitter = self.splitter
+        max_depth = self.max_depth
+        min_samples_leaf = self.min_samples_leaf
+        min_weight_leaf = self.min_weight_leaf
+        min_samples_split = self.min_samples_split
+        min_impurity_decrease = self.min_impurity_decrease
+
+        # Recursive partition (without actual recursion)
+        splitter.init(X, y, sample_weight, missing_values_in_feature_mask)
+
+        stack = []
+
+        # Push root node onto stack
+        stack.append(
+            StackRecord(
+                start=0,
+                end=splitter.n_samples,
+                depth=0,
+                parent=_TREE_UNDEFINED,
+                is_left=False,
+                impurity=INFINITY,
+                n_constant_features=0
+            )
+        )
+        weighted_n_node_samples = np.zeros(1, dtype=np.double)
+        while stack:
+            stack_record = stack.pop()
+
+            start = stack_record.start
+            end = stack_record.end
+            depth = stack_record.depth
+            parent = stack_record.parent
+            is_left = stack_record.is_left
+            impurity = stack_record.impurity
+            n_constant_features = stack_record.n_constant_features
+
+            n_node_samples = end - start
+            splitter.node_reset(start, end, weighted_n_node_samples)
+
+            is_leaf = (
+                depth >= max_depth
+                or n_node_samples < min_samples_split
+                or n_node_samples < 2 * min_samples_leaf
+                or np.sum(sample_weight[start:end]) < 2 * min_weight_leaf
+            )
+
+            if is_left:
+                impurity = splitter.node_impurity()
+
+            is_leaf = is_leaf or impurity <= EPSILON
+
+            if not is_leaf:
+                split = SplitRecord()  # No idea what is SplitRecord in original code. Maybe this never gets called, not sure
+                splitter.node_split(impurity, split, n_constant_features)
+                is_leaf = (
+                    is_leaf
+                    or split.pos >= end
+                    or (split.improvement + EPSILON < min_impurity_decrease)
+                )
+
+            node_id = tree._add_node(
+                parent,
+                is_left,
+                is_leaf,
+                split.feature if not is_leaf else 0,
+                split.threshold if not is_leaf else 0,
+                impurity,
+                n_node_samples,
+                np.sum(sample_weight[start:end]),
+                split.missing_go_to_left,
+            )
+
+            if node_id == np.iinfo(np.intp).max:
+                raise MemoryError()
+
+            splitter.node_value(tree.value + node_id * tree.value_stride)
+
+            if not is_leaf:
+                # Push right child on stack
+                stack.append(
+                    StackRecord(
+                        start=split.pos,
+                        end=end,
+                        depth=depth + 1,
+                        parent=node_id,
+                        is_left=False,
+                        impurity=split.impurity_right,
+                        n_constant_features=n_constant_features,
+                    )
+                )
+                # Push left child on stack
+                stack.append(
+                    StackRecord(
+                        start=start,
+                        end=split.pos,
+                        depth=depth + 1,
+                        parent=node_id,
+                        is_left=True,
+                        impurity=split.impurity_left,
+                        n_constant_features=n_constant_features,
+                    )
+                )
+
+
 class Tree:
     def __init__(self, n_features, n_classes, n_outputs):
         """Constructor."""
@@ -190,9 +331,7 @@ def __setstate__(self, d):
 
     def _resize(self, capacity):
         """
-        Resize all inner arrays to `capacity`.
-
-        If `capacity` is -1, then double the size of the inner arrays.
+        Resize all inner arrays to `capacity`. If `capacity` is -1, then double the size of the inner arrays.
         Returns -1 in case of failure to allocate memory (and raise MemoryError), or 0 otherwise.
         """
         if self._resize_c(capacity) != 0:
@@ -258,9 +397,7 @@ def _add_node(
         #     if self._resize_c() != 0:
         #         return -1 #throw error if resize not possible
 
-        node = (
-            Node()
-        )  # self.nodes contains a list of nodes, it returns the node at node_id location
+        node = Node() #self.nodes contains a list of nodes, it returns the node at node_id location
         self.nodes.append(node)
         node.impurity = impurity
         node.n_node_samples = n_node_samples
@@ -294,9 +431,7 @@ def predict(self, X):
         # Get the internal data as a NumPy array
         internal_data = self._get_value_ndarray()
         # Use the predictions to index the internal data
-        out = internal_data[
-            predictions
-        ]  # not sure if this accurately translates to .take(self.apply(X), axis=0, mode='clip')
+        out = internal_data[predictions] #not sure if this accurately translates to .take(self.apply(X), axis=0, mode='clip')
         # Reshape the output if the model is single-output
         if self.n_outputs == 1:
             out = out.reshape(X.shape[0], self.max_n_classes)
@@ -427,7 +562,7 @@ def _decision_path_dense(self, X):
             indices[indptr[i + 1]] = node - self.nodes
             indptr[i + 1] += 1
 
-        indices = indices[: indptr[n_samples]]
+        indices = indices[:indptr[n_samples]]
         data = np.ones(shape=len(indices), dtype=np.intp)
         out = csr_matrix((data, indices, indptr), shape=(n_samples, self.node_count))
 
@@ -488,7 +623,7 @@ def _decision_path_sparse_csr(self, X):
             indices[indptr[i + 1]] = node - self.nodes
             indptr[i + 1] += 1
 
-        indices = indices[: indptr[n_samples]]
+        indices = indices[:indptr[n_samples]]
         data = np.ones(shape=len(indices), dtype=np.intp)
         out = csr_matrix((data, indices, indptr), shape=(n_samples, self.node_count))
 
@@ -497,7 +632,6 @@ def _decision_path_sparse_csr(self, X):
     def compute_node_depths(self):
         """
         Compute the depth of each node in a tree.
-
         .. versionadded:: 1.3
 
         Returns
@@ -606,20 +740,12 @@ def compute_partial_dependence(self, X, target_features, out):
 
                 if current_node.left_child == _TREE_LEAF:
                     # Leaf node
-                    out[sample_idx] += (
-                        weight_stack[stack_size] * self.value[current_node_idx]
-                    )
+                    out[sample_idx] += weight_stack[stack_size] * self.value[current_node_idx]
                     total_weight += weight_stack[stack_size]
                 else:
-                    is_target_feature = any(
-                        target_feature == current_node.feature
-                        for target_feature in target_features
-                    )
+                    is_target_feature = any(target_feature == current_node.feature for target_feature in target_features)
                     if is_target_feature:
-                        if (
-                            X[sample_idx, current_node.feature]
-                            <= current_node.threshold
-                        ):
+                        if X[sample_idx, current_node.feature] <= current_node.threshold:
                             node_idx_stack.append(current_node.left_child)
                             weight_stack.append(weight_stack[stack_size])
                             stack_size += 1