Error running latent_ode.py about easy-neural-ode

[itamblyn]

Bump for this. Exactly same error.

[jacobjinkelly]

Hello! Sorry for the delayed reply. I'm having some trouble reproducing this error actually. I used the preprocessed data available in the release. My conda environment export is:

channels:
  - defaults
dependencies:
  - ca-certificates=2021.7.5=hecd8cb5_1
  - certifi=2021.5.30=py38hecd8cb5_0
  - libcxx=12.0.0=h2f01273_0
  - libffi=3.3=hb1e8313_2
  - ncurses=6.2=h0a44026_1
  - openssl=1.1.1l=h9ed2024_0
  - python=3.8.11=h88f2d9e_1
  - readline=8.1=h9ed2024_0
  - setuptools=58.0.4=py38hecd8cb5_0
  - sqlite=3.36.0=hce871da_0
  - tk=8.6.10=hb0a8c7a_0
  - wheel=0.37.0=pyhd3eb1b0_1
  - xz=5.2.5=h1de35cc_0
  - zlib=1.2.11=h1de35cc_3
  - pip:
    - absl-py==0.14.0
    - dm-haiku==0.0.5.dev0
    - flatbuffers==2.0
    - jax==0.2.20
    - jaxlib==0.1.71
    - jmp==0.0.2
    - numpy==1.21.2
    - opt-einsum==3.3.0
    - pip==21.2.4
    - scipy==1.7.1
    - six==1.16.0
    - tabulate==0.8.9

I ran the command python latent_ode.py --reg r2 --lam 1e-2 --test_freq 1 on my laptop and ran python latent_ode.py --reg r2 --lam 1e-2 --test_freq 1, so far I have after ~10 minutes of running on my macbook:

Iter 0001 | Loss 798.138111 | Likelihood -808.377092 | KL 2.490536 | MSE 0.165348 | Enc. r 0.000000 | Dec. r 0.001278 | Enc. NFE 0.000000 | Dec. NFE 31.824688
Iter 0002 | Loss 551.387941 | Likelihood -566.549929 | KL 1.965105 | MSE 0.116983 | Enc. r 0.000000 | Dec. r 0.005880 | Enc. NFE 0.000000 | Dec. NFE 31.839688
Iter 0003 | Loss 495.621342 | Likelihood -497.331870 | KL 1.669389 | MSE 0.103139 | Enc. r 0.000000 | Dec. r 0.020152 | Enc. NFE 0.000000 | Dec. NFE 34.642188
Iter 0004 | Loss 332.830424 | Likelihood -335.500099 | KL 1.934213 | MSE 0.070773 | Enc. r 0.000000 | Dec. r 0.016797 | Enc. NFE 0.000000 | Dec. NFE 32.999062
Iter 0005 | Loss 222.494621 | Likelihood -230.846079 | KL 2.180931 | MSE 0.049842 | Enc. r 0.000000 | Dec. r 0.010237 | Enc. NFE 0.000000 | Dec. NFE 35.735312

In particular, I used r2 since it uses less memory. Using r3 is possible, but I typically only ran this on a remote cluster where I had access to more RAM.

When you ran the first time and got the error TypeError: '<class 'jaxlib.xla_extension.DeviceArray'>' object does not support item assignment. JAX arrays are immutable; perhaps you want jax.ops.index_update or jax.ops.index_add instead?, was this after you ran the data processing code yourself?

In summary, my suggestions are:

1. See if my conda environment is different than yours and if this fixes this error.
2. Set --test_freq 1 to confirm code is running (the default is --test_freq 640
3. Try --reg r2 since it's faster and uses less memory
4. Try running on a remote machine with more RAM, especially if you want to use --reg r3, e.g. try Google Collab?

Please let me know if any of this is helpful, or if you have any other issues!