Implement a generative model for time series to model SFH

EiffL / sfh-inference-hackathon

Repository for SFH inference hackathon at AstroInfo 2021

MIT License

7 stars 2 forks source link

Implement a generative model for time series to model SFH #9

Closed EiffL closed 2 years ago

EiffL commented 2 years ago

This task is to try to implement a generative model for 1D data

Here are a few pointers:

notebook with a 1d temporal network.

EiffL commented 2 years ago

I have implemented a Keras shift layer to break the connection between present and https://colab.research.google.com/drive/1JoYHMHWvO151AW2E833uffPl7h7tz7SY?usp=sharing

nicolas-cerardi commented 2 years ago

Work in progress on pixelcnn/tfds branch. Currently the training crashes because of incompatible shape.

nicolas-cerardi commented 2 years ago

Bug fixed, we have a training now !

EiffL commented 2 years ago

Allllez!!!!

EiffL commented 2 years ago

Hereer is my notebook https://github.com/EiffL/sfh-inference-hackathon/blob/u/EiffL/pixelcnn/notebooks/PixelCNN_experiments_EiffL.ipynb

EiffL commented 2 years ago

@mhuertascompany how cool is this :-D ? SFH drawn from a pixelcnn conditioned only on SED. the curves in light orange in the background are individual posterior samples

EiffL commented 2 years ago

mhuertascompany commented 2 years ago

yeahhh!!!! so cool. let's write a paper :))

EiffL commented 2 years ago

ezgif com-gif-maker (9) ezgif com-gif-maker (8)

laurilaatu commented 2 years ago

VAE predictions don't have the fine detail of the input sample. What could be some ways to fix this issue?

sfr_vae1

EiffL commented 2 years ago

I'm going to close this issue for now so that we can start fresh