search

rapidsai / cuml

cuML - RAPIDS Machine Learning Library
https://docs.rapids.ai/api/cuml/stable/
Apache License 2.0
4.23k stars 532 forks source link

TSNE plot from cuml has tiling effect #4896

Closed Yasmin-Kassim closed 2 years ago

Yasmin-Kassim commented 2 years ago

Describe the bug I'm using cuml for tsne, and the result has some tiling effects, I've already changed and tried many parameters but I wasn't able to get a better shape, please advice.

Steps/Code to reproduce bug from cuml.manifold import TSNE adata.obsm['X_tsne'] = TSNE(perplexity=30, early_exaggeration=12, learning_rate=200).fit_transform(adata.X)

Expected behavior we should have results that doesn't have that tiling effects or at least that was what we observed in scanpy cpu tsne, however here the tsne plot has tiling effects

Environment details (please complete the following information):

cjnolet commented 2 years ago

Hi Kassim,

Thanks for opening an issue here and providing some context behind the behavior you are seeing. Since these algorithms are stochastic, there are subtle differences in the GPU TSNE implementation that can sometimes lead to subtle differences in the output when compared to Scikit-learn or Multi-core tsne.

We still often find the output is correct in many of these cases, though. It would be helpful for us to triage this issue if you are also able to provide a plot of what you are expecting to see here.

Sent from my iPhone

On Sep 22, 2022, at 4:11 PM, Yasmin M. Kassim @.***> wrote:

 Describe the bug I'm using cuml for tsne, and the result has some tiling effects, I've already changed and tried many parameters but I wasn't able to get a better shape, please advice.

Steps/Code to reproduce bug from cuml.manifold import TSNE adata.obsm['X_tsne'] = TSNE(perplexity=30, early_exaggeration=12, learning_rate=200).fit_transform(adata.X)

Expected behavior we should have results that doesn't have that tiling effects or at least that was what we observed in scanpy cpu tsne, however here the tsne plot has tiling effects

Environment details (please complete the following information):

conda enviroment on WSL GPU Model/Driver: [Quadro RTX 5000] Method of cuDF & cuML install: [conda] conda list: Name Version Build Channel

_libgcc_mutex 0.1 conda_forge conda-forge _openmp_mutex 4.5 2_gnu conda-forge abseil-cpp 20211102.0 h27087fc_1 conda-forge aiohttp 3.8.1 py39hb9d737c_1 conda-forge aiosignal 1.2.0 pyhd8ed1ab_0 conda-forge altgraph 0.17.2 pypi_0 pypi anndata 0.8.0 pypi_0 pypi anyio 3.6.1 py39hf3d152e_0 conda-forge appdirs 1.4.4 pyh9f0ad1d_0 conda-forge argon2-cffi 21.3.0 pyhd8ed1ab_0 conda-forge argon2-cffi-bindings 21.2.0 py39hb9d737c_2 conda-forge arrow-cpp 7.0.0 py39h0f417f0_8_cuda conda-forge arrow-cpp-proc 3.0.0 cuda conda-forge asttokens 2.0.5 pyhd8ed1ab_0 conda-forge async-timeout 4.0.2 pyhd8ed1ab_0 conda-forge attrs 22.1.0 pyh71513ae_1 conda-forge aws-c-cal 0.5.11 h95a6274_0 conda-forge aws-c-common 0.6.2 h7f98852_0 conda-forge aws-c-event-stream 0.2.7 h3541f99_13 conda-forge aws-c-io 0.10.5 hfb6a706_0 conda-forge aws-checksums 0.1.11 ha31a3da_7 conda-forge aws-sdk-cpp 1.8.186 hb4091e7_3 conda-forge backcall 0.2.0 pyh9f0ad1d_0 conda-forge backports 1.0 py_2 conda-forge backports.functools_lru_cache 1.6.4 pyhd8ed1ab_0 conda-forge beautifulsoup4 4.11.1 pyha770c72_0 conda-forge bleach 5.0.1 pyhd8ed1ab_0 conda-forge blosc 1.21.1 h83bc5f7_3 conda-forge bokeh 2.4.3 py39hf3d152e_0 conda-forge boost 1.74.0 py39h5472131_5 conda-forge boost-cpp 1.74.0 h312852a_4 conda-forge brotli 1.0.9 h166bdaf_7 conda-forge brotli-bin 1.0.9 h166bdaf_7 conda-forge brotlipy 0.7.0 py39hb9d737c_1004 conda-forge brunsli 0.1 h9c3ff4c_0 conda-forge bzip2 1.0.8 h7f98852_4 conda-forge c-ares 1.18.1 h7f98852_0 conda-forge c-blosc2 2.2.0 h7a311fb_0 conda-forge ca-certificates 2022.6.15 ha878542_0 conda-forge cachetools 5.0.0 pyhd8ed1ab_0 conda-forge cairo 1.16.0 h6cf1ce9_1008 conda-forge certifi 2022.6.15 py39hf3d152e_0 conda-forge cffi 1.15.1 py39he91dace_0 conda-forge cfitsio 3.470 hb418390_7 conda-forge charls 2.2.0 h9c3ff4c_0 conda-forge charset-normalizer 2.1.0 pyhd8ed1ab_0 conda-forge click 8.0.4 py39hf3d152e_0 conda-forge click-plugins 1.1.1 py_0 conda-forge cligj 0.7.2 pyhd8ed1ab_1 conda-forge cloudpickle 2.1.0 pyhd8ed1ab_0 conda-forge colorama 0.4.5 pyhd8ed1ab_0 conda-forge colorcet 3.0.0 pyhd8ed1ab_0 conda-forge cryptography 37.0.4 py39hd97740a_0 conda-forge cucim 22.06.00 cuda_11_py39_g575240a_0 rapidsai cuda-python 11.7.0 py39h3fd9d12_0 nvidia cudatoolkit 11.5.1 hcf5317a_9 nvidia cudf 22.06.01 cuda_11_py39_g0dab0f8305_0 rapidsai cudf_kafka 22.06.01 py39_g0dab0f8305_0 rapidsai cugraph 22.06.01 cuda11_py39_gde8036b5_0 rapidsai cuml 22.06.01 cuda11_py39_ge38b5bdd5_0 rapidsai cupy 9.6.0 py39ha91d172_0 conda-forge curl 7.83.1 h7bff187_0 conda-forge cusignal 22.06.00 py39_g7660efa_0 rapidsai cuspatial 22.06.00 py39_g96613ba_0 rapidsai custreamz 22.06.01 py39_g0dab0f8305_0 rapidsai cuxfilter 22.06.00 py39_gda2f3f6_0 rapidsai cycler 0.11.0 pyhd8ed1ab_0 conda-forge cyrus-sasl 2.1.27 h230043b_5 conda-forge cytoolz 0.12.0 py39hb9d737c_0 conda-forge dask 2022.5.2 pyhd8ed1ab_0 conda-forge dask-core 2022.5.2 pyhd8ed1ab_0 conda-forge dask-cuda 22.06.00 py39_0 rapidsai dask-cudf 22.06.01 cuda_11_py39_g0dab0f8305_0 rapidsai datashader 0.13.1a py_0 rapidsai datashape 0.5.4 py_1 conda-forge debugpy 1.6.0 py39h5a03fae_0 conda-forge decorator 5.1.1 pyhd8ed1ab_0 conda-forge defusedxml 0.7.1 pyhd8ed1ab_0 conda-forge distributed 2022.5.2 pyhd8ed1ab_0 conda-forge dlpack 0.5 h9c3ff4c_0 conda-forge entrypoints 0.4 pyhd8ed1ab_0 conda-forge executing 0.9.1 pyhd8ed1ab_0 conda-forge expat 2.4.8 h27087fc_0 conda-forge faiss-proc 1.0.0 cuda rapidsai fastavro 1.5.4 py39hb9d737c_0 conda-forge fastrlock 0.8 py39h5a03fae_2 conda-forge fiona 1.8.20 py39h427c1bf_2 conda-forge flit-core 3.7.1 pyhd8ed1ab_0 conda-forge font-ttf-dejavu-sans-mono 2.37 hab24e00_0 conda-forge font-ttf-inconsolata 3.000 h77eed37_0 conda-forge font-ttf-source-code-pro 2.038 h77eed37_0 conda-forge font-ttf-ubuntu 0.83 hab24e00_0 conda-forge fontconfig 2.14.0 h8e229c2_0 conda-forge fonts-conda-ecosystem 1 0 conda-forge fonts-conda-forge 1 0 conda-forge fonttools 4.34.4 py39hb9d737c_0 conda-forge freetype 2.10.4 h0708190_1 conda-forge freexl 1.0.6 h7f98852_0 conda-forge frozenlist 1.3.1 py39hb9d737c_0 conda-forge fsspec 2022.7.1 pyhd8ed1ab_0 conda-forge gdal 3.3.2 py39h218ed2d_3 conda-forge geopandas 0.9.0 pyhd8ed1ab_1 conda-forge geopandas-base 0.9.0 pyhd8ed1ab_1 conda-forge geos 3.9.1 h9c3ff4c_2 conda-forge geotiff 1.7.0 h08e826d_2 conda-forge gettext 0.19.8.1 h73d1719_1008 conda-forge gflags 2.2.2 he1b5a44_1004 conda-forge giflib 5.2.1 h36c2ea0_2 conda-forge glog 0.6.0 h6f12383_0 conda-forge grpc-cpp 1.46.3 hc275302_1 conda-forge h5py 3.7.0 pypi_0 pypi hdf4 4.2.15 h10796ff_3 conda-forge hdf5 1.12.1 nompi_h2386368_104 conda-forge heapdict 1.0.1 py_0 conda-forge holoviews 1.14.6 pyhd8ed1ab_0 conda-forge icu 68.2 h9c3ff4c_0 conda-forge idna 3.3 pyhd8ed1ab_0 conda-forge imagecodecs 2021.8.26 py39h44211f0_1 conda-forge imageio 2.21.1 pyhfa7a67d_0 conda-forge importlib-metadata 4.11.4 py39hf3d152e_0 conda-forge importlib_resources 5.9.0 pyhd8ed1ab_0 conda-forge ipykernel 6.15.1 pyh210e3f2_0 conda-forge ipython 8.4.0 py39hf3d152e_0 conda-forge ipython_genutils 0.2.0 py_1 conda-forge ipywidgets 7.7.1 pyhd8ed1ab_0 conda-forge jbig 2.1 h7f98852_2003 conda-forge jedi 0.18.1 pyhd8ed1ab_2 conda-forge jinja2 3.1.2 pyhd8ed1ab_1 conda-forge joblib 1.1.0 pyhd8ed1ab_0 conda-forge jpeg 9e h166bdaf_2 conda-forge json-c 0.15 h98cffda_0 conda-forge jsonschema 4.9.1 pyhd8ed1ab_0 conda-forge jupyter-server-proxy 3.2.1 pyhd8ed1ab_0 conda-forge jupyter_client 7.3.4 pyhd8ed1ab_0 conda-forge jupyter_core 4.11.1 py39hf3d152e_0 conda-forge jupyter_server 1.18.1 pyhd8ed1ab_0 conda-forge jupyterlab_pygments 0.2.2 pyhd8ed1ab_0 conda-forge jupyterlab_widgets 1.1.1 pyhd8ed1ab_0 conda-forge jxrlib 1.1 h7f98852_2 conda-forge kealib 1.4.15 hfe1a663_0 conda-forge keyutils 1.6.1 h166bdaf_0 conda-forge kiwisolver 1.4.4 py39hf939315_0 conda-forge krb5 1.19.3 h3790be6_0 conda-forge lcms2 2.12 hddcbb42_0 conda-forge ld_impl_linux-64 2.36.1 hea4e1c9_2 conda-forge lerc 3.0 h9c3ff4c_0 conda-forge libaec 1.0.6 h9c3ff4c_0 conda-forge libblas 3.9.0 15_linux64_openblas conda-forge libbrotlicommon 1.0.9 h166bdaf_7 conda-forge libbrotlidec 1.0.9 h166bdaf_7 conda-forge libbrotlienc 1.0.9 h166bdaf_7 conda-forge libcblas 3.9.0 15_linux64_openblas conda-forge libcrc32c 1.1.2 h9c3ff4c_0 conda-forge libcucim 22.06.00 cuda11_g575240a_0 rapidsai libcudf 22.06.01 cuda11_g0dab0f8305_0 rapidsai libcudf_kafka 22.06.01 g0dab0f8305_0 rapidsai libcugraph 22.06.01 cuda11_gde8036b5_0 rapidsai libcugraph_etl 22.06.01 cuda11_gde8036b5_0 rapidsai libcugraphops 22.06.00 cuda11_gb4e1cbb_0 nvidia libcuml 22.06.01 cuda11_ge38b5bdd5_0 rapidsai libcumlprims 22.06.00 cuda11_g8c6e2c0_0 nvidia libcurl 7.83.1 h7bff187_0 conda-forge libcusolver 11.4.0.1 0 nvidia libcuspatial 22.06.00 cuda11_g96613ba_0 rapidsai libdap4 3.20.6 hd7c4107_2 conda-forge libdeflate 1.8 h7f98852_0 conda-forge libedit 3.1.20191231 he28a2e2_2 conda-forge libev 4.33 h516909a_1 conda-forge libevent 2.1.10 h9b69904_4 conda-forge libfaiss 1.7.0 cuda112h5bea7ad_8_cuda conda-forge libffi 3.4.2 h7f98852_5 conda-forge libgcc-ng 12.1.0 h8d9b700_16 conda-forge libgcrypt 1.10.1 h166bdaf_0 conda-forge libgdal 3.3.2 h6acdded_3 conda-forge libgfortran-ng 12.1.0 h69a702a_16 conda-forge libgfortran5 12.1.0 hdcd56e2_16 conda-forge libglib 2.72.1 h2d90d5f_0 conda-forge libgomp 12.1.0 h8d9b700_16 conda-forge libgoogle-cloud 1.40.2 hefc27d0_0 conda-forge libgpg-error 1.45 hc0c96e0_0 conda-forge libgsasl 1.10.0 h5b4c23d_0 conda-forge libiconv 1.16 h516909a_0 conda-forge libkml 1.3.0 h238a007_1014 conda-forge liblapack 3.9.0 15_linux64_openblas conda-forge libllvm11 11.1.0 hf817b99_3 conda-forge libnetcdf 4.8.1 nompi_h329d8a1_102 conda-forge libnghttp2 1.47.0 h727a467_0 conda-forge libnsl 2.0.0 h7f98852_0 conda-forge libntlm 1.4 h7f98852_1002 conda-forge libopenblas 0.3.20 pthreads_h78a6416_1 conda-forge libpng 1.6.37 h753d276_3 conda-forge libpq 13.5 hd57d9b9_1 conda-forge libprotobuf 3.20.1 h6239696_0 conda-forge libraft-distance 22.06.00 cuda11_ged2c529_0 rapidsai libraft-headers 22.06.00 cuda11_ged2c529_0 rapidsai libraft-nn 22.06.00 cuda11_ged2c529_0 rapidsai librdkafka 1.7.0 hc49e61c_1 conda-forge librmm 22.06.01 cuda11_g1c7ff61a_0 rapidsai librttopo 1.1.0 h1185371_6 conda-forge libsodium 1.0.18 h36c2ea0_1 conda-forge libspatialindex 1.9.3 h9c3ff4c_4 conda-forge libspatialite 5.0.1 h5cf074c_8 conda-forge libssh2 1.10.0 ha56f1ee_2 conda-forge libstdcxx-ng 12.1.0 ha89aaad_16 conda-forge libthrift 0.16.0 h519c5ea_1 conda-forge libtiff 4.3.0 h6f004c6_2 conda-forge libutf8proc 2.7.0 h7f98852_0 conda-forge libuuid 2.32.1 h7f98852_1000 conda-forge libuv 1.43.0 h7f98852_0 conda-forge libwebp 1.2.2 h3452ae3_0 conda-forge libwebp-base 1.2.2 h7f98852_1 conda-forge libxcb 1.13 h7f98852_1004 conda-forge libxgboost 1.6.0dev.rapidsai22.06 cuda_11_0 rapidsai libxml2 2.9.12 h72842e0_0 conda-forge libzip 1.9.2 hc869a4a_0 conda-forge libzlib 1.2.12 h166bdaf_2 conda-forge libzopfli 1.0.3 h9c3ff4c_0 conda-forge llvmlite 0.38.1 py39h7d9a04d_0 conda-forge locket 1.0.0 pyhd8ed1ab_0 conda-forge lxml 4.9.1 pypi_0 pypi lz4 4.0.0 py39h029007f_2 conda-forge lz4-c 1.9.3 h9c3ff4c_1 conda-forge mapclassify 2.4.3 pyhd8ed1ab_0 conda-forge markdown 3.4.1 pyhd8ed1ab_0 conda-forge markupsafe 2.1.1 py39hb9d737c_1 conda-forge matplotlib-base 3.5.2 py39h700656a_1 conda-forge matplotlib-inline 0.1.3 pyhd8ed1ab_0 conda-forge mistune 0.8.4 py39h3811e60_1005 conda-forge msgpack-python 1.0.4 py39hf939315_0 conda-forge multidict 6.0.2 py39hb9d737c_1 conda-forge multipledispatch 0.6.0 py_0 conda-forge munch 2.5.0 py_0 conda-forge munkres 1.1.4 pyh9f0ad1d_0 conda-forge natsort 8.1.0 pypi_0 pypi nbclient 0.6.6 pyhd8ed1ab_0 conda-forge nbconvert 6.5.0 pyhd8ed1ab_0 conda-forge nbconvert-core 6.5.0 pyhd8ed1ab_0 conda-forge nbconvert-pandoc 6.5.0 pyhd8ed1ab_0 conda-forge nbformat 5.4.0 pyhd8ed1ab_0 conda-forge nccl 2.13.4.1 h0800d71_0 conda-forge ncurses 6.3 h27087fc_1 conda-forge nest-asyncio 1.5.5 pyhd8ed1ab_0 conda-forge networkx 2.6.3 pyhd8ed1ab_1 conda-forge nodejs 14.18.3 h92b4a50_1 conda-forge notebook 6.4.12 pyha770c72_0 conda-forge nspr 4.32 h9c3ff4c_1 conda-forge nss 3.78 h2350873_0 conda-forge numba 0.55.2 py39h66db6d7_0 conda-forge numpy 1.22.4 py39hc58783e_0 conda-forge nvtx 0.2.3 py39h3811e60_1 conda-forge openjpeg 2.4.0 hb52868f_1 conda-forge openssl 1.1.1q h166bdaf_0 conda-forge orc 1.7.5 h6c59b99_0 conda-forge packaging 21.3 pyhd8ed1ab_0 conda-forge pandas 1.4.3 py39h1832856_0 conda-forge pandoc 2.19 ha770c72_0 conda-forge pandocfilters 1.5.0 pyhd8ed1ab_0 conda-forge panel 0.12.7 pyhd8ed1ab_0 conda-forge param 1.12.2 pyh6c4a22f_0 conda-forge parquet-cpp 1.5.1 2 conda-forge parso 0.8.3 pyhd8ed1ab_0 conda-forge partd 1.2.0 pyhd8ed1ab_0 conda-forge patsy 0.5.2 pypi_0 pypi pcre 8.45 h9c3ff4c_0 conda-forge pexpect 4.8.0 pyh9f0ad1d_2 conda-forge pickleshare 0.7.5 py_1003 conda-forge pillow 9.1.1 py39hae2aec6_1 conda-forge pip 22.2.2 pyhd8ed1ab_0 conda-forge pixman 0.40.0 h36c2ea0_0 conda-forge pkgutil-resolve-name 1.3.10 pyhd8ed1ab_0 conda-forge poppler 21.09.0 ha39eefc_3 conda-forge poppler-data 0.4.11 hd8ed1ab_0 conda-forge postgresql 13.5 h2510834_1 conda-forge proj 8.1.0 h277dcde_1 conda-forge prometheus_client 0.14.1 pyhd8ed1ab_0 conda-forge prompt-toolkit 3.0.30 pyha770c72_0 conda-forge protobuf 3.20.1 py39h5a03fae_0 conda-forge psutil 5.9.1 py39hb9d737c_0 conda-forge pthread-stubs 0.4 h36c2ea0_1001 conda-forge ptxcompiler 0.2.0 py39h107f55c_0 rapidsai ptyprocess 0.7.0 pyhd3deb0d_0 conda-forge pure_eval 0.2.2 pyhd8ed1ab_0 conda-forge py-xgboost 1.6.0dev.rapidsai22.06 cuda_11_py39_0 rapidsai pyarrow 7.0.0 py39h1ed2e5d_8_cuda conda-forge pycparser 2.21 pyhd8ed1ab_0 conda-forge pyct 0.4.6 py_0 conda-forge pyct-core 0.4.6 py_0 conda-forge pydeck 0.5.0 pyh9f0ad1d_0 conda-forge pyee 8.1.0 pyhd8ed1ab_0 conda-forge pygments 2.12.0 pyhd8ed1ab_0 conda-forge pyinstaller 5.3 pypi_0 pypi pyinstaller-hooks-contrib 2022.8 pypi_0 pypi pylibcugraph 22.06.01 cuda11_py39_gde8036b5_0 rapidsai pynndescent 0.5.7 pypi_0 pypi pynvml 11.4.1 pyhd8ed1ab_0 conda-forge pyopenssl 22.0.0 pyhd8ed1ab_0 conda-forge pyparsing 3.0.9 pyhd8ed1ab_0 conda-forge pyppeteer 1.0.2 pyhd8ed1ab_0 conda-forge pyproj 3.1.0 py39h1fd097d_4 conda-forge pyraft 22.06.00 cuda11_py39_ged2c529_0 rapidsai pyrsistent 0.18.1 py39hb9d737c_1 conda-forge pysocks 1.7.1 py39hf3d152e_5 conda-forge python 3.9.13 h9a8a25e_0_cpython conda-forge python-confluent-kafka 1.7.0 py39h3811e60_2 conda-forge python-dateutil 2.8.2 pyhd8ed1ab_0 conda-forge python-fastjsonschema 2.16.1 pyhd8ed1ab_0 conda-forge python-pptx 0.6.21 pypi_0 pypi python_abi 3.9 2_cp39 conda-forge pytz 2022.1 pyhd8ed1ab_0 conda-forge pyviz_comms 2.2.0 pyhd8ed1ab_0 conda-forge pywavelets 1.3.0 py39hd257fcd_1 conda-forge pyyaml 6.0 py39hb9d737c_4 conda-forge pyzmq 23.2.0 py39headdf64_0 conda-forge rapids 22.06.00 cuda11_py39_ged167d1_108 rapidsai rapids-xgboost 22.06.00 cuda11_py39_ged167d1_108 rapidsai re2 2022.04.01 h27087fc_0 conda-forge readline 8.1.2 h0f457ee_0 conda-forge requests 2.28.1 pyhd8ed1ab_0 conda-forge rmm 22.06.01 cuda11_py39_g1c7ff61a_0 rapidsai rtree 1.0.0 py39hb102c33_1 conda-forge s2n 1.0.10 h9b69904_0 conda-forge scanpy 1.9.1 pypi_0 pypi scikit-image 0.19.3 py39h1832856_0 conda-forge scikit-learn 1.1.2 py39he5e8d7e_0 conda-forge scipy 1.9.0 py39h8ba3f38_0 conda-forge seaborn 0.11.2 pypi_0 pypi send2trash 1.8.0 pyhd8ed1ab_0 conda-forge session-info 1.0.0 pypi_0 pypi setuptools 60.10.0 py39hf3d152e_0 conda-forge shapely 1.8.0 py39ha61afbd_0 conda-forge simpervisor 0.4 pyhd8ed1ab_0 conda-forge six 1.16.0 pyh6c4a22f_0 conda-forge snappy 1.1.9 hbd366e4_1 conda-forge sniffio 1.2.0 py39hf3d152e_3 conda-forge sortedcontainers 2.4.0 pyhd8ed1ab_0 conda-forge soupsieve 2.3.2.post1 pyhd8ed1ab_0 conda-forge spdlog 1.8.5 h4bd325d_1 conda-forge sqlite 3.39.2 h4ff8645_0 conda-forge stack_data 0.3.0 pyhd8ed1ab_0 conda-forge statsmodels 0.13.2 pypi_0 pypi stdlib-list 0.8.0 pypi_0 pypi streamz 0.6.4 pyh6c4a22f_0 conda-forge tblib 1.7.0 pyhd8ed1ab_0 conda-forge terminado 0.15.0 py39hf3d152e_0 conda-forge threadpoolctl 3.1.0 pyh8a188c0_0 conda-forge tifffile 2021.11.2 pyhd8ed1ab_0 conda-forge tiledb 2.3.4 he87e0bf_0 conda-forge tinycss2 1.1.1 pyhd8ed1ab_0 conda-forge tk 8.6.12 h27826a3_0 conda-forge toolz 0.12.0 pyhd8ed1ab_0 conda-forge tornado 6.1 py39hb9d737c_3 conda-forge tqdm 4.64.0 pyhd8ed1ab_0 conda-forge traitlets 5.3.0 pyhd8ed1ab_0 conda-forge treelite 2.4.0 py39heaea588_0 conda-forge treelite-runtime 2.4.0 pypi_0 pypi typing-extensions 4.3.0 hd8ed1ab_0 conda-forge typing_extensions 4.3.0 pyha770c72_0 conda-forge tzcode 2022a h166bdaf_0 conda-forge tzdata 2022a h191b570_0 conda-forge ucx 1.12.1 h7a399c7_1 conda-forge ucx-proc 1.0.0 gpu rapidsai ucx-py 0.26.00 py39_g85a4428_0 rapidsai umap-learn 0.5.3 pypi_0 pypi unicodedata2 14.0.0 py39hb9d737c_1 conda-forge urllib3 1.26.11 pyhd8ed1ab_0 conda-forge wcwidth 0.2.5 pyh9f0ad1d_2 conda-forge webencodings 0.5.1 py_1 conda-forge websocket-client 1.3.3 pyhd8ed1ab_0 conda-forge websockets 10.3 py39hb9d737c_0 conda-forge wheel 0.37.1 pyhd8ed1ab_0 conda-forge widgetsnbextension 3.6.1 pyha770c72_0 conda-forge xarray 2022.6.0 pyhd8ed1ab_1 conda-forge xerces-c 3.2.3 h9d8b166_3 conda-forge xgboost 1.6.0dev.rapidsai22.06 cuda_11_py39_0 rapidsai xlsxwriter 3.0.3 pypi_0 pypi xorg-kbproto 1.0.7 h7f98852_1002 conda-forge xorg-libice 1.0.10 h7f98852_0 conda-forge xorg-libsm 1.2.3 hd9c2040_1000 conda-forge xorg-libx11 1.7.2 h7f98852_0 conda-forge xorg-libxau 1.0.9 h7f98852_0 conda-forge xorg-libxdmcp 1.1.3 h7f98852_0 conda-forge xorg-libxext 1.3.4 h7f98852_1 conda-forge xorg-libxrender 0.9.10 h7f98852_1003 conda-forge xorg-renderproto 0.11.1 h7f98852_1002 conda-forge xorg-xextproto 7.3.0 h7f98852_1002 conda-forge xorg-xproto 7.0.31 h7f98852_1007 conda-forge xz 5.2.5 h516909a_1 conda-forge yaml 0.2.5 h7f98852_2 conda-forge yarl 1.7.2 py39hb9d737c_2 conda-forge zeromq 4.3.4 h9c3ff4c_1 conda-forge zfp 0.5.5 h9c3ff4c_8 conda-forge zict 2.2.0 pyhd8ed1ab_0 conda-forge zipp 3.8.1 pyhd8ed1ab_0 conda-forge zlib 1.2.12 h166bdaf_2 conda-forge zlib-ng 2.0.6 h166bdaf_0 conda-forge zstd 1.5.2 h8a70e8d_3 conda-forge Additional context Add any other context about the problem here.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.

bencheikhb1 commented 2 years ago

Hi Corey,

Thanks for your quick reply. We agree that the output is overall correct. The issue described here by Yasmin is about the tiling effects. As in the figure that she attached, we can see blocks (squares) of dots if you zoom in on the plot. Which is not expected. And we see this in all our samples. (one sample is usually ~300K to ~3 million cells).

Although these tiling effects don't affect the analysis, they are aesthetically unpleasant.

Thank you for the great work. It was really helpful for us.

Best, Bassem

cjnolet commented 2 years ago

@bencheikhb1 thanks for the additional info. I was taking tiling effects to mean the separation between the clusters but I now see what you mean in the image. Indeed, that does seem really strange. By comparison, here's an image generated from cuml's TSNE from one of our example single-cell analysis notebooks, which at first glance doesn't appear to have the tiling, though it's not zoomed in so maybe it does still have the issue.

image

  1. Does this always seem to happen with the same dataset or have you noticed it happening across different datasets?
  2. If possible, can you try using method='barnes_hut' and let us know if you still observe the same effects?
bencheikhb1 commented 2 years ago

@cjnolet The example you sent is using Scanpy with Multicore Fast TSNE, not CUML: sc.tl.tsne(adata, n_pcs=tsne_n_pcs, use_fast_tsne=True)

In our case, we are using the CUML version on GPU: from cuml.manifold import TSNE adata.obsm['X_tsne'] = TSNE(perplexity=30, early_exaggeration=12, learning_rate=200).fit_transform(adata.X)

This is the notebook we followed: https://github.com/NVIDIA-Genomics-Research/rapids-single-cell-examples/blob/master/notebooks/hlca_lung_gpu_analysis.ipynb

We don't have a problem with the Scanpy version, Our tiling issue is only seen with the GPU CUML version.

Thanks

cjnolet commented 2 years ago

Doh! I indeed copied the wrong link and image in my previous response. @bencheikhb1, I re-ran the CPU and GPU notebooks again with the parameters you provided. I also tried running both Barnes-but and FFT methods on the GPU notebook, but I'm still seeing quite a close match between CPU and GPU and I'm not noticing the tiling artifacts.

It will help to know what dataset produced the example image above. Was the image produced on one of the sample datasets in the rapids-single-cell notebooks or was it produced on a different dataset? Do you notice this issue on multiple datasets or just one?

As mentioned in my response above, the cuml TSNE estimator has an argument to change the underlying algorithm which is used. By default it uses the fft algorithm. Can you try `method='barnes_hut' and see if it's producing this image?

If zooming in to the image is required to see this issue, can you provide a code snippet for us to reproduce?

Yasmin-Kassim commented 2 years ago

Hi, thanks a lot for suggesting the 'barnes_hut' method, the problem is solved, and our tSNE plot is fine now.