Error while using Natural Gradient

[MichaelVitali]

Hi, I'm using NGBoost for a project and I've just installed the last versions of numpy, pandas. I already worked with the library and I never had problems; however now I'm encountering a problem with the natural gradient. I get the following error:

Traceback (most recent call last):
  File "/Users/michaelvitali/repos/predico/main.py", line 5, in <module>
    ngb_model.main()
    ~~~~~~~~~~~~~~^^
  File "/Users/michaelvitali/repos/predico/models/ngb_model.py", line 109, in main
    model.fit(X_train_sc, y_train_sc.flatten())
    ~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/envs/predico/lib/python3.13/site-packages/ngboost/ngboost.py", line 236, in fit
    grads = D.grad(Y_batch, natural=self.natural_gradient)
  File "/opt/anaconda3/envs/predico/lib/python3.13/site-packages/ngboost/scores.py", line 12, in grad
    grad = np.linalg.solve(metric, grad)
  File "/opt/anaconda3/envs/predico/lib/python3.13/site-packages/numpy/linalg/_linalg.py", line 410, in solve
    r = gufunc(a, b, signature=signature)
ValueError: solve: Input operand 1 has a mismatch in its core dimension 0, with gufunc signature (m,m),(m,n)->(m,n) (size 14400 is different from 2).

On the other side, when setting natural_gradient=False works.
Do you know how can I solve it?

Thank you in advance.

[ryan-wolbeck]

Thanks for reporting this issue. To help us replicate the error and rule out any environment-related causes, could you please provide additional details such as:
• The exact versions of Python, numpy, pandas, and any other dependencies you’re using.
• Your operating system and installation method (e.g., pip or conda).
• A minimal reproducible example or detailed steps to replicate the error.

Also, @alejandroschuler , could you share your thoughts on whether this might be an issue with the natural gradient math itself, or if it could be solely related to environment discrepancies?

Thanks!

[alejandroschuler]

Yeah not sure off the top of my head, would need a reproducible example

[MichaelVitali]

Hi, thanks for your answer. Here all the details:

• python version: 3.13.2
• pandas version: 2.2.3
• numpy version: 2.2.4
• ngboost version: 0.3.6
• Operating system: macOs 15.4
• installation method: conda

Here an example of code that generates the error:

X = pd.DataFrame(data=[[1,2], [1,2], [1,2]], columns=["f1", "f2"])
y = pd.Series(data=[1,1,1])

NGBRegressor().fit(X, y)

Furthermore, with this small example, when setting natural_gradient=False, it returns this error: ValueError: Input y contains NaN.