TSNE freezes on certain data[QST]

[zyh3826]

What is your question? like question #3360 my dataset

Steps/Code to reproduce

import numpy as np
from cuml import TSNE

vectors = np.load(your_path)
tsne = TSNE(n_components=2, init='random', random_state=42, perplexity=30)
Y = tsne.fit_transform(vectors)

Expected behavior

Similar size dataset usually completes in less than 10 sec.

Environment details (please complete the following information):

• Environment location: Docker
• Linux Distro/Architecture: ubuntu18.04
• GPU Model/Driver: RTX 2080Ti
• CUDA: 10.2
• Method of cuDF & cuML install: Container derived from rapidsai/rapidsai:0.19-cuda10.2-runtime-ubuntu18.04

[lowener]

Hello, I noticed that you are using cuml version 0.19 which is more than one year old. Can you try the latest version? We updated TSNE to use FFT by default, which should not hang indefinitely.

Here is how to get the latest stable docker container of cuml:

docker pull rapidsai/rapidsai-core:22.06-cuda11.5-base-ubuntu20.04-py3.9
docker run --gpus all --rm -it \
    --shm-size=1g --ulimit memlock=-1 \
    rapidsai/rapidsai-core:22.06-cuda11.5-base-ubuntu20.04-py3.9

Or you can select yourself the configuration of the docker container here in Step 3: https://rapids.ai/start.html

[zyh3826]

Hello, I noticed that you are using cuml version 0.19 which is more than one year old. Can you try the latest version? We updated TSNE to use FFT by default, which should not hang indefinitely.

Here is how to get the latest stable docker container of cuml:

docker pull rapidsai/rapidsai-core:22.06-cuda11.5-base-ubuntu20.04-py3.9
docker run --gpus all --rm -it \
    --shm-size=1g --ulimit memlock=-1 \
    rapidsai/rapidsai-core:22.06-cuda11.5-base-ubuntu20.04-py3.9

Or you can select yourself the configuration of the docker container here in Step 3: https://rapids.ai/start.html

thanks for your reply, I will try it

[github-actions[bot]]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.