Prepare kinematic data and time to last merger as tf dataset

[ppfn]

Commit 122bab3d35264ae9d346d26e97af3fd00632a81a in new Branch mergers_kin_dataset add a package for galaxy mergers dataset using kinetic data

[ppfn]

Merge pull request #11 (commit c4e719d89c9e8d4e290cf489e9f7683c80f5f735) add the kinetic module into the sfh.datasets.mergers package.

[ppfn]

@Siouar, can you work on generating the full dataset on Jean-Zay ? If I remember well, last time we tried we had connection issues and the process was interrupted.

To do so I think a simple python script containing:

import tensorflow as tf
import tensorflow_datasets as tfds
from sfh.datasets.mergers import kinetic

ds = tfds.load('MergersKinetic', split='train')

should be launched from a screen session on one of the login node of Jean-Zay.

[EiffL]

I'm running this dataset in full at the moment, and will place it in a common directory for all of us to easily use.

[ppfn]

Commit a9f30bddac1b8fc89f877902b24ed151df3bb3e6 add different BUILDER_CONFIGS to the dataset. That way, one can call the tfds.load() function with different sizes for the dataset:

ds = tfds.load('MergersKinetic/tiny') # max 250 examples in the dataset
ds = tfds.load('MergersKinetic/small') # max 2,500 examples in the dataset
ds = tfds.load('MergersKinetic/medium') # max 5,000 examples in the dataset
ds = tfds.load('MergersKinetic/full') # max 1,000,000 examples in the dataset (arbitrary large value)
ds = tfds.load('MergersKinetic') # max 1,000,000 examples in the dataset by default (full)

[ppfn]

Created a new branch variant_kinetic_dataset where kinetic.py includes BUILDER_CONFIGS for the MergersKinetic dataset are present

[EiffL]

I thinink this is all done! Thanks guys!