James Adams

Rather than using a [Conv2D](https://keras.io/layers/convolutional/#conv2d) layer for the CNN part of the model we may want to explore using a [Cropping3D](https://www.tensorflow.org/api_docs/python/tf/keras/layers/Cropping3D) layer instead.

Examples: - [pure scikit-learn](http://queirozf.com/entries/scikit-learn-pipeline-examples) - [wrapping Keras models for inclusion in a Pipeline](https://stackoverflow.com/questions/42415076/how-to-insert-keras-model-into-scikit-learn-pipeline)

Thanks, Justin! I have an example of the main direction I'm heading illustrated in a notebook, in case that'll give any guidance as to what I'm planning to first turn...

Of late I don't use or maintain this package. If you can work out the issue and fix the code with a PR it'll be welcome. I apologize I can't...

This is not supported yet, but I assume it is possible.

Thank you for this initial review @NickleDave . I will further document the code with docstrings and generate API docs via `autodoc`.

Yes, David. Thanks! I will be on PTO at the end of this month and hopefully, I will get to that as well as some other issues during that time....

Thanks @usmansall Can you please post a link to the dataset `alina_precipitation_data.nc` used in the example above so I can attempt to replicate the issue?

Using the latest data you posted as input and using the latest branch from this repository I was able to produce SPI datasets without an error: ``` $ git clone...

I was able to reproduce the error. The issue is that we expect the data to use (lat, lon, time) as the dimension order, and this data had the time...