`Failed` doesn't return even when input data contains `NaN` or `INFINITY` (for LinearRegression)

[lamrongol]

I noticed LinearRegression::fit() result parameters( such as intercept(), coefficients()) are all NaN when input data(arguments of fit() ) contains NaN or INFINITY.
I think it should return Failed.

I confirmed this problem only for LinearRegression, but other modules may have same problem.

[Mec-iS]

yes. There is no handling of NaN, every dataset should be assumed finite or make it so before being passed to any construct in smartcore; for performance reasons, most of the algorithms do not scan the entire dataset (as iterators are used anywhere possible) so it is not possible to say if there are any non-finite values before-hand. In theory failing when a NaN value is encountered in the values should be the right behaviour but again checking every value of a dataset for finiteness will take too much overhead.
Maybe a good option is to check at the end if the results are not finite and panic printing out something like "your results are non-finite values, it is possible that there were non-finite values in your dataset".
For a backgrond on design decisions: #185

[lamrongol]

for performance reasons, most of the algorithms do not scan the entire dataset
Maybe a good option is to check at the end if the results are not finite and panic printing out something like "your results are non-finite values, it is possible that there were non-finite values in your dataset".

I see. I think check at the end is good idea.