upgrade pfns to latest

nanotabpfn/interface.py

@@ -5,7 +5,7 @@
 import requests
 import torch
 import torch.nn.functional as F
-from pfns.bar_distribution import FullSupportBarDistribution
+from pfns.model.bar_distribution import FullSupportBarDistribution
 from sklearn.compose import ColumnTransformer
 from sklearn.impute import SimpleImputer
 from sklearn.pipeline import Pipeline

nanotabpfn/train.py

@@ -3,7 +3,7 @@
 import time
 from torch.utils.data import DataLoader
 from typing import Dict
-from pfns.bar_distribution import FullSupportBarDistribution
+from pfns.model.bar_distribution import FullSupportBarDistribution
 import schedulefree
 import os
 

nanotabpfn/utils.py

@@ -3,7 +3,7 @@
 import torch
 import numpy as np
 
-from pfns.bar_distribution import get_bucket_limits
+from pfns.model.bar_distribution import get_bucket_borders
 
 def set_randomness_seed(seed):
     random.seed(seed)
@@ -32,5 +32,5 @@ def make_global_bucket_edges(filename, n_buckets=100, device=get_default_device(
         raise ValueError(f"Too few target samples ({ys_concat.size}) to compute {n_buckets} buckets.")
 
     ys_tensor = torch.tensor(ys_concat, dtype=torch.float32, device=device)
-    global_bucket_edges = get_bucket_limits(n_buckets, ys=ys_tensor).to(device)
+    global_bucket_edges = get_bucket_borders(n_buckets, ys=ys_tensor).to(device)
     return global_bucket_edges

pretrain_regression.py

@@ -1,7 +1,7 @@
 import argparse
 
 import torch
-from pfns.bar_distribution import FullSupportBarDistribution
+from pfns.model.bar_distribution import FullSupportBarDistribution
 from sklearn.metrics import r2_score
 
 from nanotabpfn.callbacks import ConsoleLoggerCallback

pyproject.toml

@@ -22,7 +22,7 @@ dependencies = [
     "requests>=2",
     "scikit-learn>=1.5",
     "schedulefree>=1.4",
-    "pfns==0.3.0",
+    "pfns==0.4.2",
     "openml==0.15.1",
 ]