Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Pytorch #14

Open
wants to merge 109 commits into
base: master
Choose a base branch
from
Open

Pytorch #14

wants to merge 109 commits into from

Conversation

justinalsing
Copy link
Owner

No description provided.

justinalsing and others added 30 commits January 31, 2020 11:09
@justinalsing
updated ndes.py with tf2 MDN (no MAF or MADE yet)
2b04615
@justinalsing
added sample() function to MDN
236e3bf
@justinalsing
added first version of MADE and MAF, needs debugging once training cl…
e564414 
…asses are added
@justinalsing
fixed name scope of MADEs in MAF
7fb0a43
@sfeeney
TF2 training module: need to integrate and test
f5e8570
@sfeeney
TF version requirement so we don't forget!
491ba5e
@justinalsing
stitched together new training and ndes classes, cleared TF1 bits (se…
9b73fa4 
…ssions etc) out of delfi.py, updated cosmic_shear.ipynb example
@justinalsing
Merge branch 'tf2' of https://github.com/justinalsing/pydelfi into tf2
83f9728
@justinalsing
fixed all tf2 NDEs including conditional MADE and MAF
e3f45ef
@justinalsing
updated getdist behaviour and way we interact with emcee
e81756f
@justinalsing
changed emcee initialization based on top nwalkers samples
4fbf071
@justinalsing
updated example notebooks to work with new interface, expect priors. …
3e9f58c 
…changed emcee initialization to randomly sample previous posterior samples from all but the bottom third of log-posterior values
@justinalsing
adjusted emcee initialization scheme again, cutting off bottom 10% of…
90c1ec4 
… samples only
@justinalsing
updated all priors in examples to tfd, updated versions of requiremen…
b148b88 
…ts in setup.py, updated saver() function in delfi.py
@justinalsing
fixed restore capabilities in saver() function in delfi.py
9b091ba
MAF in tfp working (cannot yet sample) - working on efficient trainin…
6de4847 
…g routine (nearly done)
@justinalsing
fixed triangle plot to use weights if passes and changed tight_layout…
dd4f304 
… to bbox_inches=tight inside savefig()
Working upto fisher pretraining. Need to change proposal in sequentia…
3d474cf 
…l to tfp (might already be done on tf2 branch). Early stopping has a bug. Don't know if can install directly from pip yet (pulling tfp from my repo). Need to decide on weighting - I've gone with using -lnL, and the weighting log_prob, but really we should use exp(-lnL) and prob. The problem with this is that it can set all weights to zero when distribution if far from ideal. Really we should maximise the ELBO rather than doing what we are doing right now...
Working delfi
bb8fbb2
typo in setup.py
852be74
Changed some float value issues
1f60ca5
removed priors.py
e49fb70
TF2 update (done?)
49001af
@justinalsing
changed loss to look like KL divergence (reduce_sum changed to reduce…
43992fb 
…_mean)
@justinalsing
fixed size of training/validation sets with tf.random.shuffle instead…
7fee5cd 
… of catgorical
@justinalsing
fixed batching of training and val data to put them on same scale
aee37a5
@justinalsing
fixed number of sig figs on tqdm outputs
f704179
@justinalsing
fixed number of sig figs on tqdm outputs attempt 2
00e0990
@justinalsing
made MDN full rank
2b43839
@justinalsing
re-introduced normalization into training data and thetas
ce60638
justinalsing and others added 30 commits November 5, 2020 11:34
@justinalsing
fixed saver() in sequential training and filename path bug
ac8860f
@justinalsing
killed buggy axis label rotaiton on getdist plots
8cbec89
@justinalsing
added sinharcsinh mades
909df87
@justinalsing
added conditional functions to sinh MADEs
07e278e
@justinalsing
initialize variables in sinh made
947e7b7
@justinalsing
changed optimizer defaults
9e8b680
@justinalsing
replaced emcee with tf affine sampler
ed5eb49
@justinalsing
replaced emcee with tf affine sampler, fixed import
bfc1ac9
@justinalsing
replaced emcee with tf affine sampler, fixed self bug
9d85dcf
@justinalsing
replaced emcee with tf affine sampler, fixed self bug in fisher pretr…
60bccbd 
…aining
@justinalsing
replaced emcee with tf affine sampler, fixed n_walkers typo
16cd49d
@justinalsing
fixed sampling progress bar
80ac9d3
@justinalsing
fixed sampling progress bar take II
7a21e4c
@justinalsing
removed posterior weighting for triangle plots
9d6a6b9
@justinalsing
added burn in steps
5d38a34
@justinalsing
changed x0 initialization back
a3a56af
@justinalsing
fixed shapes of stack probabilities
6a2fdcc
@justinalsing
fixed shapes of stack probabilities take II
673b51e
@justinalsing
fixed shapes of stack probabilities take III
60de9e4
@justinalsing
fixed shapes of stack probabilities take IV
5d3cd6c
@justinalsing
testing @tf.function behaviour of sinharcsinh made
0b871b8
@justinalsing
changed weighted prob to logsumexp
1534546
@justinalsing
purged some python natives
c626071
@justinalsing
purged some python natives II
ef473fe
@justinalsing
clipped output of affine sampler
c4cfe22
@justinalsing
added weighting of affine samples for GetDist
217e81d
@justinalsing
added weighting of affine samples for GetDist II
314029e
@justinalsing
updated examples
6d95161
@NiallJeffrey
tf 2.4 use
5ebb0ef
@justinalsing
removed event shape bug in MAFs
ddc98b5
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@justinalsing @sfeeney @NiallJeffrey