Spaces:
Running
Running
File size: 7,768 Bytes
e17f3ba e057d08 e17f3ba e057d08 e17f3ba e057d08 e17f3ba e057d08 fc6a2fc e057d08 fc6a2fc e057d08 e17f3ba | 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 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 | """
TabPFN Wrapper
==============
Sklearn-compatible wrapper for TabPFN (Tabular Pre-trained Transformers).
TabPFN is a pretrained model for tabular classification using
in-context learning (no training required).
Author: UW MSIM Team
Date: November 2025
"""
import time
import logging
import os
from typing import Optional, Union
import numpy as np
import pandas as pd
# Automatically accept the TabPFN license to prevent browser/socket crashes on Windows
os.environ["TABPFN_LICENSE"] = "accept"
os.environ["TABPFN_ACCEPT_LICENSE"] = "1"
# ── Patch for old TabPFN compatibility with newer torch ──────────────────────
try:
import torch.nn.modules.transformer
if not hasattr(torch.nn.modules.transformer, 'Optional'):
import typing
torch.nn.modules.transformer.Optional = typing.Optional
torch.nn.modules.transformer.Any = typing.Any
torch.nn.modules.transformer.Tuple = typing.Tuple
torch.nn.modules.transformer.List = typing.List
except (ImportError, AttributeError):
pass
# ── Patch for old TabPFN compatibility with newer sklearn ────────────────────
try:
import sklearn.utils.validation
def _patch_validation(func):
from functools import wraps
@wraps(func)
def wrapper(*args, **kwargs):
if 'force_all_finite' in kwargs:
kwargs['ensure_all_finite'] = kwargs.pop('force_all_finite')
return func(*args, **kwargs)
return wrapper
sklearn.utils.validation.check_X_y = _patch_validation(sklearn.utils.validation.check_X_y)
sklearn.utils.validation.check_array = _patch_validation(sklearn.utils.validation.check_array)
except (ImportError, AttributeError):
pass
from .base_wrapper import BaseModelWrapper
logger = logging.getLogger(__name__)
class TabPFNWrapper(BaseModelWrapper):
"""
TabPFN (Tabular Prior-Fitted Networks) wrapper.
TabPFN uses pretrained transformers for zero-shot tabular prediction.
Works best on datasets with <1000 samples and <100 features.
Parameters
----------
task_type : str, default='classification'
Task type (only 'classification' supported by TabPFN)
n_ensemble : int, default=1
Number of ensemble members
device : str, default='auto'
Device: 'cpu', 'cuda', or 'auto'
random_state : int, default=42
Random seed
"""
def __init__(
self,
task_type: str = 'classification',
n_ensemble: int = 1,
device: str = 'auto',
random_state: int = 42
):
super().__init__(task_type=task_type, random_state=random_state)
if task_type != 'classification':
raise ValueError("TabPFN only supports classification tasks")
self.n_ensemble = n_ensemble
self.device = device
def fit(self, X: Union[pd.DataFrame, np.ndarray], y: Union[pd.Series, np.ndarray]) -> 'TabPFNWrapper':
"""
Fit TabPFN (stores training data for in-context learning).
Parameters
----------
X : pd.DataFrame or np.ndarray, shape (n_samples, n_features)
Training features (max 1000 samples, 100 features)
y : pd.Series or np.ndarray, shape (n_samples,)
Training target
Returns
-------
self : TabPFNWrapper
Fitted model
"""
self._validate_input(X, y)
# Check TabPFN constraints
if X.shape[0] > 1024:
logger.warning(f"TabPFN strictly requires <= 1024 samples to avoid Memory OOM. Subsampling {X.shape[0]} to 1024 samples.")
sample_idx = np.random.RandomState(self.random_state).choice(
len(X), 1024, replace=False
)
if isinstance(X, pd.DataFrame):
X = X.iloc[sample_idx]
else:
X = X[sample_idx]
if isinstance(y, pd.Series):
y = y.iloc[sample_idx]
else:
y = y[sample_idx]
if X.shape[1] > 100:
logger.warning(f"TabPFN strictly requires <= 100 features. Truncating {X.shape[1]} to 100 features.")
if isinstance(X, pd.DataFrame):
X = X.iloc[:, :100]
else:
X = X[:, :100]
self.truncated_features_ = True
else:
self.truncated_features_ = False
logger.info(f"Fitting TabPFN on {X.shape[0]} samples...")
start_time = time.time()
try:
from tabpfn import TabPFNClassifier
import torch
import tabpfn
actual_device = 'cuda' if (self.device == 'auto' and torch.cuda.is_available()) else ('cpu' if self.device == 'auto' else self.device)
if hasattr(tabpfn, '__version__') and tabpfn.__version__.startswith('0.1'):
self.model = TabPFNClassifier(device=actual_device, N_ensemble_configurations=self.n_ensemble)
else:
self.model = TabPFNClassifier(device=actual_device)
# Fit model
self.model.fit(X, y)
self.is_fitted = True
self.fit_time = time.time() - start_time
logger.info(f"TabPFN fitted in {self.fit_time:.2f} seconds")
except ImportError:
logger.error("TabPFN not installed")
raise ImportError("Install TabPFN with: pip install tabpfn")
except Exception as e:
logger.error(f"Error fitting TabPFN: {e}")
raise
return self
def predict(self, X: Union[pd.DataFrame, np.ndarray]) -> np.ndarray:
"""
Make predictions with TabPFN.
Parameters
----------
X : pd.DataFrame or np.ndarray, shape (n_samples, n_features)
Test features
Returns
-------
predictions : np.ndarray, shape (n_samples,)
Predicted class labels
"""
if not self.is_fitted:
raise ValueError("Model not fitted. Call fit() first.")
self._validate_input(X)
if getattr(self, 'truncated_features_', False) and X.shape[1] > 100:
if isinstance(X, pd.DataFrame):
X = X.iloc[:, :100]
else:
X = X[:, :100]
logger.info(f"Predicting on {X.shape[0]} samples with TabPFN...")
start_time = time.time()
try:
predictions = self.model.predict(X)
self.predict_time = time.time() - start_time
logger.info(f"Predictions complete in {self.predict_time:.2f} seconds")
return predictions
except Exception as e:
logger.error(f"Error during prediction: {e}")
raise
def _predict_proba_impl(self, X: Union[pd.DataFrame, np.ndarray]) -> np.ndarray:
"""
Predict class probabilities with TabPFN.
Parameters
----------
X : pd.DataFrame or np.ndarray, shape (n_samples, n_features)
Test features
Returns
-------
probabilities : np.ndarray, shape (n_samples, n_classes)
Class probabilities
"""
if getattr(self, 'truncated_features_', False) and X.shape[1] > 100:
if isinstance(X, pd.DataFrame):
X = X.iloc[:, :100]
else:
X = X[:, :100]
return self.model.predict_proba(X)
def get_params(self, deep: bool = True) -> dict:
"""Get parameters for this estimator."""
params = super().get_params(deep)
params.update({
'n_ensemble': self.n_ensemble,
'device': self.device
})
return params
|