AutoTabPFNRegressor: ValueError: Buffer dtype mismatch, expected 'const float32_t' but got 'double'

[Mccti078]

Describe the bug

Hi Tabpfn,

I'm trying to use AutoTabPFNRegressor to train a tabular model. I keep getting the below error. I can't seem to fix it. For context, I have no problem using TabPFNRegressor for the same dataset.

My apologies if this is user error, but I am at a bit of a dead end.

Thank you

2025-02-05 20:44:50 INFO Using default preset for Post Hoc Ensemble.
2025-02-05 20:44:50 INFO Using categorical_feature_indices: [2]
2025-02-05 20:44:50 INFO Using task type: regression
2025-02-05 20:44:50 INFO Obtaining TabPFN models from a random portfolio.
2025-02-05 20:44:52 INFO Using 100 base models: ['default_tabpfn_model_0', 'random_tabpfn_model_1', 'random_rf_pfn_model_2', 'random_tabpfn_model_3', 'random_rf_pfn_model_4', 'random_rf_pfn_model_5', 'random_rf_pfn_model_6', 'random_rf_pfn_model_7', 'random_tabpfn_model_8', 'random_tabpfn_model_9', 'random_rf_pfn_model_10', 'random_rf_pfn_model_11', 'random_tabpfn_model_12', 'random_rf_pfn_model_13', 'random_tabpfn_model_14', 'random_tabpfn_model_15', 'random_rf_pfn_model_16', 'random_tabpfn_model_17', 'random_tabpfn_model_18', 'random_rf_pfn_model_19', 'random_rf_pfn_model_20', 'random_rf_pfn_model_21', 'random_tabpfn_model_22', 'random_rf_pfn_model_23', 'random_tabpfn_model_24', 'random_tabpfn_model_25', 'random_rf_pfn_model_26', 'random_tabpfn_model_27', 'random_tabpfn_model_28', 'random_rf_pfn_model_29', 'random_tabpfn_model_30', 'random_rf_pfn_model_31', 'random_tabpfn_model_32', 'random_tabpfn_model_33', 'random_rf_pfn_model_34', 'random_tabpfn_model_35', 'random_rf_pfn_model_36', 'random_rf_pfn_model_37', 'random_tabpfn_model_38', 'random_rf_pfn_model_39', 'random_tabpfn_model_40', 'random_tabpfn_model_41', 'random_tabpfn_model_42', 'random_rf_pfn_model_43', 'random_tabpfn_model_44', 'random_tabpfn_model_45', 'random_rf_pfn_model_46', 'random_tabpfn_model_47', 'random_rf_pfn_model_48', 'random_tabpfn_model_49', 'random_tabpfn_model_50', 'random_tabpfn_model_51', 'random_tabpfn_model_52', 'random_rf_pfn_model_53', 'random_tabpfn_model_54', 'random_tabpfn_model_55', 'random_rf_pfn_model_56', 'random_rf_pfn_model_57', 'random_rf_pfn_model_58', 'random_tabpfn_model_59', 'random_rf_pfn_model_60', 'random_tabpfn_model_61', 'random_rf_pfn_model_62', 'random_rf_pfn_model_63', 'random_tabpfn_model_64', 'random_rf_pfn_model_65', 'random_rf_pfn_model_66', 'random_rf_pfn_model_67', 'random_tabpfn_model_68', 'random_tabpfn_model_69', 'random_tabpfn_model_70', 'random_rf_pfn_model_71', 'random_tabpfn_model_72', 'random_rf_pfn_model_73', 'random_rf_pfn_model_74', 'random_rf_pfn_model_75', 'random_rf_pfn_model_76', 'random_rf_pfn_model_77', 'random_rf_pfn_model_78', 'random_tabpfn_model_79', 'random_rf_pfn_model_80', 'random_tabpfn_model_81', 'random_tabpfn_model_82', 'random_tabpfn_model_83', 'random_tabpfn_model_84', 'random_tabpfn_model_85', 'random_tabpfn_model_86', 'random_rf_pfn_model_87', 'random_rf_pfn_model_88', 'random_tabpfn_model_89', 'random_tabpfn_model_90', 'random_rf_pfn_model_91', 'random_rf_pfn_model_92', 'random_rf_pfn_model_93', 'random_tabpfn_model_94', 'random_tabpfn_model_95', 'random_tabpfn_model_96', 'random_tabpfn_model_97', 'random_rf_pfn_model_98', 'random_rf_pfn_model_99']
2025-02-05 20:44:52 INFO Starting 80-repeated holdout validation with holdout_frac=0.33.
2025-02-05 20:44:52 INFO Set time limit to 2500 seconds. We will early stop validation if needed.
2025-02-05 20:44:52 INFO Yield data for model default_tabpfn_model_0 and split 0 (repeat=1).
2025-02-05 20:45:43 INFO Yield data for model random_tabpfn_model_1 and split 0 (repeat=1).
2025-02-05 20:46:00 INFO Yield data for model random_rf_pfn_model_2 and split 0 (repeat=1).
2025-02-05 20:46:00 INFO Using default preset for Post Hoc Ensemble.
2025-02-05 20:46:00 INFO Using categorical_feature_indices: [2]
2025-02-05 20:46:00 INFO Using task type: regression
2025-02-05 20:46:00 INFO Obtaining TabPFN models from a random portfolio.

An error occurred: Buffer dtype mismatch, expected 'const float32_t' but got 'double'

Full traceback:
Traceback (most recent call last):
File "C:\Users\User\AppData\Local\Temp\ipykernel_6384\507384520.py", line 75, in run_tabpfn_tuned
model.fit(X_train_np, y_train_np)
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\sklearn_interface.py", line 222, in fit
self.predictor_.fit(
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\pfn_phe.py", line 333, in fit
self._ens_model.fit(X, y)
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\greedy_weighted_ensemble.py", line 234, in fit
weights = self.get_weights(X, y)
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\greedy_weighted_ensemble.py", line 173, in get_weights
oof_proba = self.get_oof_per_estimator(X, y)
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\abstract_validation_utils.py", line 372, in get_oof_per_estimator
self._fill_predictions_in_place(
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\post_hoc_ensembles\abstract_validation_utils.py", line 127, in _fill_predictions_in_place
base_model.fit(fold_X_train, fold_y_train)
File "C:\Users\User\tabpfn-extensions\src\tabpfn_extensions\rf_pfn\SklearnBasedRandomForestTabPFN.py", line 98, in fit
super().fit(X, y)
File "C:\Users\User\anaconda3\envs\ml_testing\lib\site-packages\sklearn\base.py", line 1351, in wrapper
return fit_method(estimator, *args, **kwargs)
File "C:\Users\User\anaconda3\envs\ml_testing\lib\site-packages\sklearn\ensemble_forest.py", line 377, in fit
estimator._compute_missing_values_in_feature_mask(
File "C:\Users\User\anaconda3\envs\ml_testing\lib\site-packages\sklearn\tree_classes.py", line 228, in _compute_missing_values_in_feature_mask
missing_values_in_feature_mask = _any_isnan_axis0(X)
File "sklearn\tree\_utils.pyx", line 450, in sklearn.tree._utils._any_isnan_axis0
ValueError: Buffer dtype mismatch, expected 'const float32_t' but got 'double'

Steps/Code to Reproduce

No response

Expected Results

No response

Actual Results

No response

Versions

[LeoGrin]

Hi @Mccti078 !
Thanks for the report :) Would you be able to share a minimal reproducible example?

[Mccti078]

Hello @LeoGrin yes absolutely. I have a jupyter notebook and a dataset I can share with you. Perhaps I could email it to you?

[LeoGrin]

Hey @Mccti078, thanks a lot! This should actually be fixed by PriorLabs/tabpfn-community#23, could you upgrade your tabpfn-extension package and try again?