Random forests don't support nans, in fit or in predict. Is this a feature request or a bug report? The HistGradientBoosting esitmators are the only ones to natively support nans (I think). You'll need impute the data if you want to use RFs.

The error message you get is strange. How did you fit the estimator? You should get something like

ValueError: Expected 2D array, got 1D array instead:
array=[nan nan nan nan nan nan nan nan nan nan].
Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample.

X is expected to be (n_samples, n_features) and there is not support for missing values in random forest. We add some work going on supporting missing values natively in decision tree in scikit-learn: #5974

I am closing this issue because it looks like a duplicate.

@MQSchleich Feel free to comment in case that we missed something in your original, in which case we would reopen this issue.

Random forests don't support nans, in fit or in predict. Is this a feature request or a bug report? The HistGradientBoosting esitmators are the only ones to natively support nans (I think). You'll need impute the data if you want to use RFs.

The error message you get is strange. How did you fit the estimator? You should get something like

ValueError: Expected 2D array, got 1D array instead:
array=[nan nan nan nan nan nan nan nan nan nan].
Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample.

Honestly, I was just reproducing a case without giving you the whole code base. The bug occurred when getting input from a data frame after converting it to a list, having 1 np.nan value. However, it is not bad at all that the forest throws an error. Only the message was so confusing, I needed some time to figure out what was going on. Therefore, I suggest it is actually better to not throw an error but continue processing and to throw a warning.