-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathxgboost.py
61 lines (50 loc) · 2.08 KB
/
xgboost.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
import typing as tp
from leo_transpiler.boostings.core import BoostingTranspiler
from leo_transpiler.leo import LeoIfElseNode, LeoNode, LeoReturnNode
from leo_transpiler.quantize import quantize
class XgboostTranspiler(BoostingTranspiler):
"""
Transpiler for XGBoost models.
"""
def __init__(self, model, quantize_bits: int = 8):
"""
:param model: The XGBoost model.
:param quantize_bits: The number of bits to quantize to.
"""
trees = model.get_booster()
super().__init__(
model=model,
feature_names=trees.feature_names,
n_classes=hasattr(model, "n_classes_") and model.n_classes_ or None,
n_estimators=model.n_estimators,
quantize_bits=quantize_bits
)
self._dfs = []
for i in range(self.n_estimators):
df = trees[i].trees_to_dataframe()
if self.is_regression:
self._dfs.append(df)
else:
for c in range(self.n_classes):
class_df = df[df["Tree"] == c].reset_index(drop=True)
self._dfs.append(class_df)
def get_leo_ast_nodes(self) -> tp.List[LeoNode]:
return [self.build_tree(df, df.iloc[0]) for df in self._dfs]
def build_tree(self, df, df_node) -> LeoNode:
feature_name = df_node["Feature"]
if feature_name != "Leaf":
if_node_id = df_node["Yes"]
else_node_id = df_node["No"]
if_node = df.iloc[self.node_id_to_idx(if_node_id)]
else_node = df.iloc[self.node_id_to_idx(else_node_id)]
value = quantize(df_node["Split"], self.quantize_bits)
condition = f"{feature_name} < {value}"
left = self.build_tree(df, if_node)
right = self.build_tree(df, else_node)
return LeoIfElseNode(condition, left, right)
else:
value = quantize(df_node["Gain"], self.quantize_bits)
return LeoReturnNode(value)
@staticmethod
def node_id_to_idx(node_id: str) -> int:
return int(node_id.split("-")[1])