3D interactive session using phoenix did not show any content

[ZhenHuangLab]

Issue Type

bug

MiLoPYP version

v0.5.0

Description

When initiating the 3D interactive session locally in Cellular Content Exploration by running the code below:

python phoenix_visualization.py --input exp/simsiam2d3d/test_sample/interactive_info_parquet.gzip

The session was hosted successfully, and by opening the web address I could see the title Phoenix and its icon. But, it did not show any content and was always blank. I have waited for hours and tried refreshing or re-launching the phoenix but they didn't work.

Do you have any suggestions?

Additional Info

• I have followed the instructions in Quick tutorials for globular-shaped targets and the dataset is from milopyp_globular_tutorial.tbz. I just paused at 3D interactive session section.
• The 2D visualization was successful.
• The images generated by plot_2d.py are hosted in the background on port 7000 and they are accessible.

GPU

Tesla V100-PCIE-16GB * 1

CUDA version

cuda-11.7

Software Setup

torch==1.13.0+cu117 torchaudio==0.13.0 torchvision==0.14.0+cu117
arize-phoenix 0.0.27, matplotlib 3.7.5, numpy 1.24.4, numba 0.58.1, scipy 1.10.1

[abartesaghi]

This seems to be an issue with the Phoenix app itself. Did you try clearing the cache or using a different browser? Here is a discussion on a similar issue that could be helpful.

[ZhenHuangLab]

I found that cleaning my browser cache didn't work, as mentioned in the similar issue you provided.

Then I tried setting up nvm and pnpm to build the phoenix app (using node v16.20.2 and npm v8.19.4 because glibc in centos7 doesn't support node 18 and higher, and it's kind of annoying to upgrade glibc), by following the instruction in phoenix/app/readme and phoenix/development:

$ nvm use 16
Now using node v16.20.2 (npm v8.19.4)

# now in phoenix/app/
$ pnpm install --frozen-lockfile
Lockfile is up to date, resolution step is skipped
Already up to date
Done in 21.9s

$ pip show arize-phoenix
Name: arize-phoenix
Version: 5.6.0
Summary: AI Observability and Evaluation
Home-page:
Author:
Author-email: Arize AI <phoenix-devs@arize.com>
License: Elastic-2.0
Location: .../lab01/conda_env/envs/phoenix/lib/python3.9/site-packages
Editable project location: .../zhen/phoenix
Requires: aioitertools, aiosqlite, alembic, arize-phoenix-evals, arize-phoenix-otel, authlib, cachetools, fastapi, grpc-interceptor, grpcio, httpx, jinja2, numpy, openinference-instrumentation, openinference-semantic-conventions, opentelemetry-exporter-otlp, opentelemetry-proto, opentelemetry-sdk, opentelemetry-semantic-conventions, pandas, protobuf, psutil, pyarrow, pydantic, python-multipart, scikit-learn, scipy, sqlalchemy, sqlean-py, starlette, strawberry-graphql, tqdm, typing-extensions, uvicorn, websockets, wrapt
Required-by:

$ pnpm run build

> phoenix-ui@0.0.1 build .../zhen/phoenix/app
> pnpm run build:relay && vite build


> phoenix-ui@0.0.1 build:relay .../zhen/phoenix/app
> relay-compiler

[INFO] [default] compiling...
[INFO] [default] compiled documents: 139 reader, 104 normalization, 149 operation text
[INFO] Compilation completed.
[INFO] Done.
vite v5.4.10 building for production...
node_modules/.pnpm/three-stdlib@2.34.0_three@0.139.2/node_modules/three-stdlib/libs/lottie.js (13062:32): Use of eval in "node_modules/.pnpm/three-stdlib@2.34.0_three@0.139.2/node_modules/three-stdlib/libs/lottie.js" is strongly discouraged as it poses security risks and may cause issues with minification.
node_modules/.pnpm/three-mesh-bvh@0.7.8_three@0.139.2/node_modules/three-mesh-bvh/src/utils/ExtensionUtilities.js (6:26): "BatchedMesh" is not exported by "node_modules/.pnpm/three@0.139.2/node_modules/three/build/three.module.js", imported by "node_modules/.pnpm/three-mesh-bvh@0.7.8_three@0.139.2/node_modules/three-mesh-bvh/src/utils/ExtensionUtilities.js".
✓ 4117 modules transformed.
../../src/phoenix/server/static/.vite/manifest.json                        1.93 kB │ gzip:     0.40 kB
../../src/phoenix/server/static/assets/vendor-Cg6lcjUC.css                 1.82 kB │ gzip:     0.77 kB
../../src/phoenix/server/static/assets/index-DqLgwdks.js                   7.16 kB │ gzip:     2.41 kB
../../src/phoenix/server/static/assets/vendor-recharts-DgwO5t0F.js       298.80 kB │ gzip:    67.10 kB
../../src/phoenix/server/static/assets/components-v7Tbuo8y.js            299.91 kB │ gzip:    82.30 kB
../../src/phoenix/server/static/assets/vendor-arizeai-D373FxfT.js        307.00 kB │ gzip:    84.67 kB
../../src/phoenix/server/static/assets/vendor-codemirror-B-gpTz_s.js     393.56 kB │ gzip:   127.20 kB
../../src/phoenix/server/static/assets/pages-DULbjTlg.js                 578.18 kB │ gzip:   114.82 kB
../../src/phoenix/server/static/assets/vendor-three-t78-ZqWt.js          620.97 kB │ gzip:   158.16 kB
../../src/phoenix/server/static/assets/vendor-AVwlOmm_.js             10,942.71 kB │ gzip: 2,224.72 kB

(!) Some chunks are larger than 500 kB after minification. Consider:
- Using dynamic import() to code-split the application
- Use build.rollupOptions.output.manualChunks to improve chunking: https://rollupjs.org/configuration-options/#output-manualchunks
- Adjust chunk size limit for this warning via build.chunkSizeWarningLimit.
✓ built in 55.40s

The phoenix app seemed to be successfully built. But, when I tried to run the command once again, it crashed again:

# Now in cet_pick/
$ python phoenix_visualization.py --input exp/simsiam2d3d/test_sample/interactive_info_parquet.gzip
Traceback (most recent call last):
  File ".../zhen/MiLoPYP/cet_pick/cet_pick/phoenix_visualization.py", line 2, in <module>
    import phoenix as px
  File ".../zhen/phoenix/src/phoenix/__init__.py", line 10, in <module>
    from .session.client import Client
  File ".../zhen/phoenix/src/phoenix/session/client.py", line 33, in <module>
    from phoenix.experiments.types import Dataset, Example, Experiment
  File ".../zhen/phoenix/src/phoenix/experiments/__init__.py", line 1, in <module>
    from .functions import evaluate_experiment, run_experiment
  File ".../zhen/phoenix/src/phoenix/experiments/functions.py", line 34, in <module>
    from phoenix.evals.executors import get_executor_on_sync_context
ModuleNotFoundError: No module named 'phoenix.evals'

Then I re-installed the arize-phoenix by pip install arize-phoenix[evals], and revised the phoenix_visualization.py because in the latest arize-phoenix (v5.6.0) px.Dataset was renamed to px.Inferences:

# in phoenix_visualization.py
- train_ds = px.Dataset(dataframe=train_df, schema=train_schema)
+ train_ds = px.Inferences(dataframe=train_df, schema=train_schema)

It seemed that it finally worked. But I could not open http://localhost:6006/:

$ python phoenix_visualization.py --input exp/simsiam2d3d/test_sample/interactive_info_parquet.gzip &
[1] 28027
name            coord  ... label                             image
0  tilt1   [266, 97, 126]  ...    38  http://localhost:7000/imgs/0.png
1  tilt1   [355, 76, 134]  ...    32  http://localhost:7000/imgs/1.png
2  tilt1  [251, 249, 124]  ...    38  http://localhost:7000/imgs/2.png
3  tilt1  [281, 273, 133]  ...    38  http://localhost:7000/imgs/3.png
4  tilt1   [303, 94, 133]  ...    32  http://localhost:7000/imgs/4.png

[5 rows x 5 columns]
🌍 To view the Phoenix app in your browser, visit http://localhost:6006/
📖 For more information on how to use Phoenix, check out https://docs.arize.com/phoenix

Then I undid the revision in phoenix_visualization.py, and re-installed the arize-phoenix==0.0.27.

$ python phoenix_visualization.py --input exp/simsiam2d3d/test_sample/interactive_info_parquet.gzip
.../lab01/conda_env/envs/phoenix/lib/python3.9/site-packages/pydantic/_migration.py:290: UserWarning: `pydantic.utils:lenient_issubclass` has been removed. We are importing from `pydantic.v1.utils:lenient_issubclass` instead.See the migration guide for more details: https://docs.pydantic.dev/latest/migration/
  warnings.warn(
.../lab01/conda_env/envs/phoenix/lib/python3.9/site-packages/pydantic/_migration.py:290: UserWarning: `pydantic.utils:smart_deepcopy` has been removed. We are importing from `pydantic.v1.utils:smart_deepcopy` instead.See the migration guide for more details: https://docs.pydantic.dev/latest/migration/
  warnings.warn(
    name            coord  ... label                             image
0  tilt1   [266, 97, 126]  ...    38  http://localhost:7000/imgs/0.png
1  tilt1   [355, 76, 134]  ...    32  http://localhost:7000/imgs/1.png
2  tilt1  [251, 249, 124]  ...    38  http://localhost:7000/imgs/2.png
3  tilt1  [281, 273, 133]  ...    38  http://localhost:7000/imgs/3.png
4  tilt1   [303, 94, 133]  ...    32  http://localhost:7000/imgs/4.png

[5 rows x 5 columns]
🌍 To view the Phoenix app in your browser, visit http://localhost:39301/
📺 To view the Phoenix app in a notebook, run `px.active_session().view()`
📖 For more information on how to use Phoenix, check out https://docs.arize.com/phoenix

But, the dashboard was blank again. lol

So, next step, I will try running docker on my mac, to see if it works.

Have you ever tried phoenix_visualization.py on your remote machine/cluster (in documentation it seems that you run 3D interactive session on personal laptop)? What would you suggest? Thank you!

Additional info

System: CentOS Linux 7.5.1804
Required library versions: (for phoenix-visualization)

dataset==1.6.2
healpy==1.14.0
opencv_contrib_python==4.7.0.72
Pillow==10.0.0
numba==0.60.0
numpy==1.26.4
pandas==2.2.1
arize-phoenix==0.0.27
arize-phoenix-evals==0.17.2
arize-phoenix-otel==0.5.1
pyarrow==15.0.0
scipy==1.13.1
strawberry-graphql==0.178.0
sqlean.py==3.45.1

[ZhenHuangLab]

So Strange -- when I installed arize-phoenix on my mac using these commands below:

conda create --name milopyp python=3.8
conda activate milopyp
pip install pandas==2.0.3 arize-phoenix==0.0.27
/Users/zhenhuang/opt/anaconda3/envs/milopyp/bin/python /Users/zhenhuang/Documents/milopyp/phoenix_visualization.py --input exp/simsiam2d3d/test_sample/interactive_info_parquet.gzip

I got the expected output:

And finally, I can see the dashboard in the browser, Hooray!

So, I think it should be highlighted in the documentation that it is better to run the entire 3D interactive session on one's local machine like mac, but not on the cluster or the remote machine. It's kind of confusing because all the steps before 3D interactive session could run on the remote machine.

And my questions remain:

Have you ever tried phoenix_visualization.py on your remote machine/cluster (in documentation it seems that you run 3D interactive session on personal laptop)? What would you suggest? Thank you!

It would be better if the 3D interactive session could also run on the cluster, so that there is no need to copy the interactive_info_parquet.gzip, exp/,and phoenix_visualization.py to my mac and do this stuff locally.

[huangqinwen]

Thank you! We will look into this. Last time when I tried to run it on a remote cluster it worked fine but will retest!