Create unet_lstm.py

[Aakanksha-Rana]

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Summary

This model supports 3D time-series processing.

Checklist

• I have added tests to cover my changes
• I have updated documentation (if necessary)

Acknowledgment

• I acknowledge that this contribution will be available under the Apache 2 license.

[satra]

there are some lines with todo related to a max pooling layer.

also it would be helpful to know why input shapes are flipped in the creation of data vs model. i.e. why can't they be the same. in the code there seems to be a lot of repeats. could that be compressed through some loops and parameters or partial funcs?

docstring for parameters, reference to paper/original model.

[Aakanksha-Rana]

there are some lines with todo related to a max pooling layer.

also it would be helpful to know why input shapes are flipped in the creation of data vs model. i.e. why can't they be the same. in the code there seems to be a lot of repeats. could that be compressed through some loops and parameters or partial funcs?

docstring for parameters, reference to paper/original model.

• Tensorflow doesn't provide 4D pooling/upsampling layers support yet, might need to wait till they provide the support for 4D layer or if someone wants to add that to nobrainer would be great (lets put an issue maybe for that?)
• It is not flipped. the input size is (32, 32, 32, 32). other thing is just showcasing the batch size or class during initialization.
• Yeah it can be compressed.
• I will added the docstring.