search

RoryAtBar / Abaqus_bayesian_matflow

A bayesian inference workflow using matflow
0 stars 0 forks source link

Run updated scripts on CSF3 #20

Open gcapes opened 3 weeks ago

gcapes commented 3 weeks ago

@RoryAtBar has got a new approach for this work.

  1. I want to change the GP from fitting individual data points to fitting basis functions using scikit-fda
  2. Randomly separate out training data and validation data and test the fit of the validation data (about 20% of the samples to be used not for conditioning the GP, but for checking that the predicted values fit correctly)
  3. Automatically check which of four kernels fits best rather than picking the first one that fits at all
gcapes commented 1 week ago

@RoryAtBar, PYMC_GPflow_CSF3sampler_step1.sh is giving the following error. Any ideas on this? I'm working on the gp_kernel_tester branch.

ValueError: Dimensions must be equal, but are 6 and 2 for '{{node truediv_2}} = RealDiv[T=DT_DOUBLE](strided_slice_8, truediv_2/softplus_CONSTRUCTED_AT_top_level/forward/IdentityN)' with input shapes: [1,6], [2].
RoryAtBar commented 1 week ago

Hi Gerard, This has hopefully been fixed. Changed the input dimensions for the compiled functions in the saved tensorflow model to [1, number of input dimensions]

gcapes commented 1 week ago

I guess this is the same type of problem (from PYMC_GPflow_CSF3sampler_step2.sh?

ValueError: operands could not be broadcast together with shapes (6,) (2,)
RoryAtBar commented 1 week ago

Would you mind sending the whole error message?

RoryAtBar commented 1 week ago

I forgot that I copied this script from a version of this workflow that modifies constitutive parameters (of which there were 6). I have now updated this so there are only 2 (friction and conductance)

gcapes commented 1 week ago

Thanks! Step2 looks to be running ok now (job array hasn't finished yet but no errors on those which have finished 😄)

RoryAtBar commented 1 week ago

Great! Hopefully it would have errored by now

Sent from Outlook for Androidhttps://aka.ms/AAb9ysg

From: Gerard Capes @.> Sent: Friday, September 6, 2024 4:08:42 PM To: RoryAtBar/Abaqus_bayesian_matflow @.> Cc: RoryAtBar @.>; Mention @.> Subject: Re: [RoryAtBar/Abaqus_bayesian_matflow] Run updated scripts on CSF3 (Issue #20)

Thanks! Step2 looks to be running ok now (job array hasn't finished yet but no errors on those which have finished 😄)

— Reply to this email directly, view it on GitHubhttps://github.com/RoryAtBar/Abaqus_bayesian_matflow/issues/20#issuecomment-2334267879, or unsubscribehttps://github.com/notifications/unsubscribe-auth/A3IND55INB3PU2DWRAPKQEDZVHAPVAVCNFSM6AAAAABM5YGI6KVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMZUGI3DOOBXHE. You are receiving this because you were mentioned.Message ID: @.***>

gcapes commented 5 days ago

The last 5 jobs from the job array are still running...no errors yet, but no output either. I think these need to finish before I can test step3, so this issue is blocked for now.

gcapes commented 4 days ago

These seem to be taking way longer than the others in the job array, which is suspicious.

RoryAtBar commented 4 days ago

Odd, when these jobs run for this long I get segmentation errors.

RoryAtBar commented 4 days ago

Step 3 may still work if you go to the job script and remove the line #$ -hold_jid PYMC_GPflow_CSF3sampler_step2.sh If any of the chains have completed, you should have files called Idatachain{number}.nc it will still find and agglomerate them as it uses the glob module to find them.