search

wilsonrljr / sysidentpy

A Python Package For System Identification Using NARMAX Models
https://sysidentpy.org
BSD 3-Clause "New" or "Revised" License
393 stars 78 forks source link

Support for training from multiple datasets #50

Open jtylka opened 3 years ago

jtylka commented 3 years ago

as far as I can tell, the model.fit(X,y) function can only support a single training dataset. I would like to train a model based on multiple recorded datasets (say, 5 recordings, each 10 minutes long, but not captured consecutively, so concatenation is not correct). is this possible? is this planned for a future release? see for example: https://www.mathworks.com/help/ident/ug/dealing-with-multi-experiment-data-and-merging-models.html

Gabo-Tor commented 2 years ago

This feature will be greatly appreciated. In the same way, it would be great to also use many datasets for validation. Great project by the way!

wilsonrljr commented 1 year ago

Hey! I'll release a new update next week and then I'll take a look at this feature. I already have some tests in mind, so I'll let you know as soon as I start to work on it

paulogonc commented 8 months ago

I would like to ask, if there is any update on this feature? Maybe it would be possible to provide some hints how this can be done with certain modifications on the current code basis to achieve a kind of batch run batch training.

Thanks in advance.

wilsonrljr commented 6 months ago

Hey, @paulogonc . I'll attach a notebook here showing how you can use the current features to do what the matlab multi experiment does. I haven't planned to include that as a feature in the package yet (I'm focusing on some backend changes first), but the notebook might help you

wilsonrljr commented 5 months ago

I tried to reproduce the MATLAB example using the current features of SysIdentPy. It needs validation, so I would appreciate it if someone could review this example and confirm if it makes sense.

multi_experiment.zip