Commit
·
f2b8171
1
Parent(s):
a6272af
Utilities to run app
Browse files
utils.py
ADDED
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import annotations
|
2 |
+
|
3 |
+
import re
|
4 |
+
from typing import Optional, Union
|
5 |
+
|
6 |
+
import numpy as np
|
7 |
+
import plotly.graph_objects as go
|
8 |
+
from sklearn.ensemble import GradientBoostingRegressor
|
9 |
+
|
10 |
+
class DataGenerator:
|
11 |
+
def __init__(self, formula_str: str, x_range: list, n_samples: int, seed: int) -> None:
|
12 |
+
self.formula_str = formula_str
|
13 |
+
self.x_range = x_range
|
14 |
+
self.n_samples = n_samples
|
15 |
+
self.seed = seed
|
16 |
+
self.rng = np.random.RandomState(seed)
|
17 |
+
|
18 |
+
@property
|
19 |
+
def X(self) -> np.array:
|
20 |
+
self.rng = np.random.RandomState(42)
|
21 |
+
X = np.atleast_2d(self.rng.uniform(*self.x_range, size=self.n_samples)).T
|
22 |
+
return X
|
23 |
+
|
24 |
+
@property
|
25 |
+
def y_raw(self) -> np.array:
|
26 |
+
y_raw = self._eval_formula()
|
27 |
+
return y_raw.ravel()
|
28 |
+
|
29 |
+
@property
|
30 |
+
def y(self) -> np.array:
|
31 |
+
sigma = 0.5 + self.X.ravel() / 10
|
32 |
+
noise = self.rng.lognormal(sigma=sigma) - np.exp(sigma**2 / 2)
|
33 |
+
return self.y_raw + noise
|
34 |
+
|
35 |
+
def _eval_formula(self) -> np.array:
|
36 |
+
function_map = {
|
37 |
+
'sin': "np.sin",
|
38 |
+
'cos': "np.cos",
|
39 |
+
'tan': "np.tan",
|
40 |
+
'exp': "np.exp",
|
41 |
+
'log': "np.log",
|
42 |
+
'sqrt': "np.sqrt",
|
43 |
+
'abs': "np.abs",
|
44 |
+
}
|
45 |
+
# Replace "x" in the formula string with "x_values"
|
46 |
+
_formula_str = re.sub(r'\bx\b', '(self.X)', self.formula_str)
|
47 |
+
# Replace any function calls in the formula string with the appropriate function object
|
48 |
+
_formula_str = re.sub(r'(\w+)\(([^)]*)\)', lambda m: f'{function_map[m.group(1)]}({m.group(2)})', _formula_str)
|
49 |
+
# Evaluate the formula using the updated string and return the result
|
50 |
+
return eval(_formula_str)
|
51 |
+
|
52 |
+
class GradientBoostingCoverage:
|
53 |
+
def __init__(self, lower: float, upper: float, **kwargs) -> None:
|
54 |
+
self.lower = lower
|
55 |
+
self.upper = upper
|
56 |
+
self.kwargs = kwargs
|
57 |
+
self.models = self._build_models()
|
58 |
+
|
59 |
+
@property
|
60 |
+
def expected_coverage(self) -> float:
|
61 |
+
return self.upper - self.lower
|
62 |
+
|
63 |
+
def _build_models(self) -> dict[str, GradientBoostingRegressor]:
|
64 |
+
models = {}
|
65 |
+
for name, alpha in [("lower", self.lower), ("upper", self.upper)]:
|
66 |
+
models[f"{name}"] = GradientBoostingRegressor(loss="quantile", alpha=alpha, **self.kwargs)
|
67 |
+
return models
|
68 |
+
|
69 |
+
def fit(self, X: np.ndarray, y: np.array) -> None:
|
70 |
+
for model in self.models.values():
|
71 |
+
model.fit(X, y)
|
72 |
+
|
73 |
+
def predict(self, X: np.ndarray) -> tuple[np.array, np.array]:
|
74 |
+
lower = self.models["lower"].predict(X)
|
75 |
+
upper = self.models["upper"].predict(X)
|
76 |
+
return lower, upper
|
77 |
+
|
78 |
+
def coverage_fraction(self, X: np.ndarray, y: np.array) -> float:
|
79 |
+
y_low, y_high = self.predict(X)
|
80 |
+
return np.mean(np.logical_and(y >= y_low, y <= y_high))
|
81 |
+
|
82 |
+
|
83 |
+
def fit_gradientboosting(X, y, **kwargs) -> GradientBoostingRegressor:
|
84 |
+
model = GradientBoostingRegressor(**kwargs)
|
85 |
+
model.fit(X, y)
|
86 |
+
return model
|
87 |
+
|
88 |
+
def plot_interval(
|
89 |
+
xx: np.array,
|
90 |
+
X_test: np.array,
|
91 |
+
y_test: np.array,
|
92 |
+
y_upper: np.array,
|
93 |
+
y_lower: np.array,
|
94 |
+
y_med: np.array,
|
95 |
+
y_mean: np.array,
|
96 |
+
formula_str: Optional[str]=None,
|
97 |
+
interval: Optional[Union[int, str]]=None,
|
98 |
+
) -> go.Figure:
|
99 |
+
# Using plotly to plot an interval
|
100 |
+
fig = go.Figure()
|
101 |
+
|
102 |
+
fig.add_trace(
|
103 |
+
go.Scatter(
|
104 |
+
x=xx.ravel(),
|
105 |
+
y=y_upper,
|
106 |
+
fill=None,
|
107 |
+
mode="lines",
|
108 |
+
line_color="rgba(255,255,0,0)",
|
109 |
+
name=""
|
110 |
+
)
|
111 |
+
)
|
112 |
+
|
113 |
+
fig.add_trace(
|
114 |
+
go.Scatter(
|
115 |
+
x=xx.ravel(),
|
116 |
+
y=y_lower,
|
117 |
+
fill="tonexty",
|
118 |
+
mode="lines",
|
119 |
+
line_color="rgba(255,255,0,0)",
|
120 |
+
name=f"Predicted Interval"
|
121 |
+
)
|
122 |
+
)
|
123 |
+
|
124 |
+
fig.add_trace(
|
125 |
+
go.Scatter(
|
126 |
+
x=xx.ravel(),
|
127 |
+
y=y_med,
|
128 |
+
mode="lines",
|
129 |
+
line_color="red",
|
130 |
+
name='Predicted Median',
|
131 |
+
)
|
132 |
+
)
|
133 |
+
|
134 |
+
fig.add_trace(
|
135 |
+
go.Scatter(
|
136 |
+
x=xx.ravel(),
|
137 |
+
y=y_mean,
|
138 |
+
mode="lines",
|
139 |
+
name='Predicted Mean',
|
140 |
+
line=dict(color='red', dash='dash')
|
141 |
+
)
|
142 |
+
)
|
143 |
+
|
144 |
+
fig.add_trace(
|
145 |
+
go.Scatter(
|
146 |
+
x=X_test.ravel(),
|
147 |
+
y=y_test,
|
148 |
+
mode="markers",
|
149 |
+
marker_color="blue",
|
150 |
+
name="Test Observations",
|
151 |
+
marker=dict(size=5, line=dict(width=2, color="DarkSlateGrey"))
|
152 |
+
)
|
153 |
+
)
|
154 |
+
|
155 |
+
fig.update_layout(
|
156 |
+
title=f"Predicted {interval}% Interval",
|
157 |
+
xaxis_title="x",
|
158 |
+
yaxis_title="f(x)" if not formula_str else formula_str,
|
159 |
+
height=600
|
160 |
+
)
|
161 |
+
|
162 |
+
return fig
|