v2.1.1 🌈

🧭 What's Changed

• [AL-1132] Added the ability to do inplace transforms (#1354) @AbhinavTuli
• Pytorch transform api (#1337) @AbhinavTuli
• Chunk engine and chunk refactor (#1307) @AbhinavTuli
• Fix hub list deletion (#1340) @AbhinavTuli
• Autocast empty samples (#1334) @farizrahman4u
• Fix to ensure agreement isn't reloaded for each slice of dataset (#1347) @AbhinavTuli
• added bench repo branch input, set units as default (#1348) @gautamkrishnar
• [tiny] let users load datasets from newer versions of hub (#1306) @nollied
• Fixes pytorch with text (#1342) @AbhinavTuli
• minor bug fix for the benchmark code (#1332) @gautamkrishnar
• Added statement in Version Control log when uncommitted data is present (#1328) @AbhinavTuli
• Fix "NumPy array is not writeable" (#1338) @aliubimov
• Fix leaked semaphore issue on mac (#1291) @AbhinavTuli

⚙️ Who Contributes

@AbhinavTuli, @aliubimov, @davidbuniat, @farizrahman4u, @gautamkrishnar, @mikayelh, @nollied and @tatevikh

v2.1.0 🌈

🧭 What's Changed

• Default shuffle buffer size 2GB (#1330) @aliubimov
• bug_report.md typo and README Badge fix (#1302) @qzylalala
• Bye Bye CircleCI (#1329) @farizrahman4u
• Refactor pytorch integration (#1257) @aliubimov
• Fixes slowdowns with Hub (#1304) @AbhinavTuli
• Added text signwall within Hub (#1323) @AbhinavTuli
• Readme typo (#1299) @farizrahman4u
• Update README.md (#1322) @istranic
• Fixes bug with groups on hub cloud datasets (#1321) @AbhinavTuli
• simplify Github actions (#1318) @gautamkrishnar
• Progressbar + bad transform hanging fix (#1295) @farizrahman4u
• Video (#1216) @FayazRahman
• Changing default dataset setting from public to private (#1317) @istranic
• Add GitHub actions (#1241) @araratpoghosyan
• enable str objects to be converted to tensorflow (#1309) @hoshimura
• Persistence fix (#1298) @farizrahman4u
• Fix/cloud dataset bugs (#1294) @nollied
• [tiny] fix formatting + update keypointscoco to int32 instead of float (#1272) @nollied
• Removed overwrite and public arguments from hub.load (#1275) @AbhinavTuli
• Jpeg fix issue with verify (#1297) @farizrahman4u
• hub.delete + logging fixes (#1278) @farizrahman4u
• Feature/add docker compose (#1250) @davidbuniat
• Tensor.bytes() (#1280) @farizrahman4u
• Add class_names as an option to segment_mask htype (#1281) @farizrahman4u
• Fix grayscale warnings + hub.read docstring (#1289) @farizrahman4u
• Better retries for s3 (#1283) @AbhinavTuli
• Byte compression + empty sample fix (#1293) @farizrahman4u
• Readme updates (#1265) @farizrahman4u
• refactor core/Dataset (#1284) @nollied
• Add token to hub list (#1276) @AbhinavTuli
• Add pytest fixture decorator to test_gcs_tokens (#1285) @dhiganthrao
• Progress bar for hub.eval (#1200) @farizrahman4u
• hub/compression.py dosctring update (#1279) @farizrahman4u
• Better error message for invalid dataset (#1282) @farizrahman4u

⚙️ Who Contributes

@AbhinavTuli, @FayazRahman, @aliubimov, @araratpoghosyan, @davidbuniat, @dhiganthrao, @farizrahman4u, @gautamkrishnar, @hoshimura, @istranic, @mikayelh, @nollied, @qzylalala and @tatevikh

v2.0.13 🌈

🧭 What's Changed

• added improvements and regressions as artifacts (#1255) @gautamkrishnar
• Fixes empty samples in hub compute (#1269) @AbhinavTuli
• [2 liner] TorchDataset.len (#1268) @mccrearyd
• Fix creating nested tensor from non-head node (#1261) @FayazRahman
• fix text not working for transforms (#1258) @mccrearyd
• [Tiny] Updated htype docstring for segment mask (#1264) @istranic
• Added coco_keypoint htype and changed default dtype for segment masks (#1262) @istranic
• Groups perf fix (#1212) @farizrahman4u
• Jpeg: External test images repo + small fixes (#1249) @farizrahman4u
• Json, List, Text (#1214) @farizrahman4u
• fix pytest dependency error (#1254) @mccrearyd
• PNG compression for arbitrary number of channels (#1252) @farizrahman4u
• [tiny] tqdm progress bar for pytorch (#1243) @mccrearyd

🔗 Dependency Updates

• Bump pillow from 8.2.0 to 8.3.2 in /hub/requirements (#1151) @dependabot

⚙️ Who Contributes

@AbhinavTuli, @FayazRahman, @davidbuniat, @dependabot, @farizrahman4u, @gautamkrishnar, @istranic and @mccrearyd

v2.0.12 🌈

🧭 What's Changed

• Fix segmentation fault (#1248) @aliubimov
• added pull_request_target check (#1246) @gautamkrishnar
• removed repository name from pr checks (#1245) @gautamkrishnar
• Added reporting to ingest, ingest_kaggle, and list (#1242) @istranic
• PyTorch & Tensorflow aliases for datasets (#1238) @muhdhisham
• added manual benchmarking for hub baseline (#1224) @gautamkrishnar
• Added benchmarking workflow to hub pull requests (#1218) @gautamkrishnar
• Fix/sagemaker creds (#1204) @AbhinavTuli
• Fix Getting Started link (#1237) @aliubimov
• Fix VC test (#1233) @farizrahman4u
• Add flac and wav support (#1220) @FayazRahman
• Fix pickle5 issue (#1230) @AbhinavTuli
• Fix grayscale tests (#1222) @farizrahman4u
• Fixes issues with kaggle tests after audio merge (#1231) @AbhinavTuli
• Beautify meta .json files (#1229) @jakbin
• Update README.md (#1194) @istranic
• Audio (#1207) @farizrahman4u
• Bring back "Linux Timple Test" (#1203) @Diveafall
• AL-1368 Handle grayscale image after color image (#1206) @jraman
• Added reporting for version control (#1209) @istranic
• [tiny] Update error messages (#1211) @mccrearyd
• Distributed computation with ray (#1140) @AbhinavTuli
• Revert "Adding shared memory support for Pytorch 3.6/3.7, speeding up pytorch integration" (#1210) @AbhinavTuli
• AL-1433 List compression types for hub.compression (#1208) @jraman
• Adding shared memory support for Pytorch 3.6/3.7, speeding up pytorch integration (#1196) @AbhinavTuli
• lz4 backward compatibility (#1201) @FayazRahman
• Update docs for hub.ingest (#1199) @jraman
• Removing "Linux Simple Test" from all workflows (#1198) @Diveafall
• Change lz4 implementation to numcodecs (#1197) @FayazRahman
• Revert "PyTorch & Tensorflow aliases for datasets" (#1195) @mccrearyd
• PyTorch & Tensorflow aliases for datasets (#1183) @muhdhisham

⚙️ Who Contributes

@AbhinavTuli, @Diveafall, @FayazRahman, @aliubimov, @davidbuniat, @farizrahman4u, @gautamkrishnar, @istranic, @jakbin, @jraman, @mccrearyd, @mikayelh, @muhdhisham and @tatevikh

Google Cloud Support and Hierarchical Datasets 🌈

🩰 What's New

• Added hierarchical datasets (tensor groups)
• Added support for Google Cloud Storage (path = gcs://...)
• Added version control alpha

🧭 What's Changed

• Fix version control speed for first commit (#1192) @AbhinavTuli
• Missing compressions fix (#1191) @farizrahman4u
• Tensor group fixes (#1190) @farizrahman4u
• gcs path fix (#1188) @davidbuniat
• Version Control (#1152) @AbhinavTuli
• One more jpg fix (#1186) @farizrahman4u
• fix running tests fails when no gcs creds exist (#1187) @mccrearyd
• Istranic update readme (#1185) @istranic
• GCS support (#1125) @kristinagrig06
• [Small] Fixes logging (#1153) @AbhinavTuli
• Modify dataset privacy (#1141) @kristinagrig06
• Imagenet weird jpegs fix (#1155) @farizrahman4u
• Hierarchical Tensors (#1139) @farizrahman4u
• [small] client/utils.py: Dont sys.exit on Bad request (#1154) @farizrahman4u
• Add Tensor to API reference (#1146) @kristinagrig06
• [small-Please approve] fix typo (#1144) @thisiseshan

⚙️ Who Contributes

@AbhinavTuli, @davidbuniat, @farizrahman4u, @istranic, @kristinagrig06, @mccrearyd, @tatevikh and @thisiseshan

v2.0.9 🌈

🧭 What's Changed

• Multi - SOF jpeg fix (#1134) @farizrahman4u
• Pytorch shuffling (#1122) @AbhinavTuli
• DS locking: atexit release (#1138) @farizrahman4u
• Treat size 1 arrays as scalars when casting (#1135) @farizrahman4u
• Tensor inplace ops (#1136) @farizrahman4u
• Small fix (#1133) @farizrahman4u
• Faster deserialization (#1131) @farizrahman4u
• hub.read speedup (4x) (#1126) @farizrahman4u
• Delete .from_path() method (#1128) @kristinagrig06
• [small] better error messages + fix ingest_kaggle (#1132) @mccrearyd
• DS lock fix (#1129) @farizrahman4u
• auto refinement (#1124) @thisiseshan
• Chunk-wise compression (#1093) @farizrahman4u
• Ingestion summary (#1117) @thisiseshan
• hub.read speed up (>6X) (#1120) @farizrahman4u
• Dataset locking (#1119) @farizrahman4u
• Backward compat fix (#1121) @farizrahman4u
• backwards compatibility (and CI tests!) (#1110) @mccrearyd
• fix pytorch tests chunk sizes + max_chunk_size in create_tensor (#1114) @mccrearyd
• Auto compression (#1109) @thisiseshan
• add --kaggle functionality (#1108) @thisiseshan
• Fixes an issue with pytorch old read only (#1113) @AbhinavTuli
• Add Dataset functions to docs (#1098) @kristinagrig06
• [small] hub.htypes is a list of all htypes (#1104) @mccrearyd

⚙️ Who Contributes

@AbhinavTuli, @davidbuniat, @farizrahman4u, @kristinagrig06, @mccrearyd, @tatevikh and @thisiseshan

Fixes to parallel computing 🌈

🧭 What's Changed

• Fix/num workers 0 bug (#1112) @davidbuniat
• Move hub.compute reporting to pipeline eval (#1116) @istranic
• Updates transform docstrings/variables to remove usage of the word "transform" (#1115) @AbhinavTuli
• Fixed bugout reporting paths and overreporting during Dataset initialization. (#1111) @davidbuniat
• Release/2.0.6 (#1107) @davidbuniat
• Faster chunk serialization (#1106) @AbhinavTuli
• Remove default logging code in client (#1097) @benchislett

⚙️ Who Contributes

@AbhinavTuli, @benchislett, @davidbuniat, @istranic and @mccrearyd

Refactors and Minor Updates

🩰 What's New

• Added hub.ingest for automatic creation of datasets
• Added hub.list to help users find publicly available datasets
• Lots of refactors to help developers

🧭 What's Changed

• kaggle argument fix (#1101) @thisiseshan
• Polish top directory (#1054) @kristinagrig06
• Ban dataset attributes as tensor names (#1103) @kristinagrig06
• update tensors (#1089) @mccrearyd
• More sample compressions (#1087) @farizrahman4u
• [small] all scalars have shape (1,) instead of () (#1102) @mccrearyd
• refactor input pipeline for samples (#1099) @mccrearyd
• Integrate hub auto + kaggle (#1075) @thisiseshan
• List datasets (#1048) @kristinagrig06

⚙️ Who Contributes

@AbhinavTuli, @farizrahman4u, @kristinagrig06, @mccrearyd and @thisiseshan

Adding metadata and parallel computations

🎁 What's New

• You can add metadata to datasets and tensors
• You can run computations in parallel using hub.compute
• The dataset API is updated to be more intuitive

🧭 What's Changed

• Add static dataset delete (#1060) @benchislett
• 2.0.4 release Version update (#1094) @davidbuniat
• Adding back transforms for parallel dataset uploads (#1086) @AbhinavTuli
• Release 2.0.3 (#1084) @davidbuniat
• Update code snippets (#1088) @istranic
• [refactor] encoders base class (#1082) @mccrearyd
• Alias "jpg" to "jpeg" (#1073) @benchislett
• Updated readme (#1085) @istranic
• [small] implement hub.like in new api (#1083) @mccrearyd
• Bugout reporting update (#1081) @istranic
• Info fixes (#1080) @farizrahman4u
• BUGGER_OFF=true when running tests in Circle CI. (#1079) @zomglings
• [small] Remove dataset link during creation of Hub datasets (#1078) @dhiganthrao
• Enable search in pdoc (#1074) @benchislett
• Added "dataset" class for interacting with underlying "Dataset" class (#1063) @AbhinavTuli
• [small] turn off activeloop reporting during circleci tests (#1076) @mccrearyd
• dataset/tensor info alongside meta (#1066) @mccrearyd
• [small] fix hub cloud throttling for tests (#1077) @mccrearyd
• Add back the history of master into main (#1061) @benchislett
• [Small PR] Removes tfds tests (#1070) @AbhinavTuli
• Old pytorch multiprocessing bug fix (#1068) @AbhinavTuli
• [Small PR] Renames hub.load to hub.read (#1064) @AbhinavTuli
• Made Dataset and LRUCache objects pickleable (#1049) @AbhinavTuli
• Alternate fix for tensor creation bug (#1065) @AbhinavTuli

⚙️ Who Contributes

@AbhinavTuli, @benchislett, @davidbuniat, @dhiganthrao, @farizrahman4u, @istranic, @mccrearyd, @tatevikh and @zomglings

Bug fix for .pytorch DataLoaders 🌈

🎁 What's New

• We mostly focused on refactoring and minor bugs.
• .pytorch() now works with pubic datasets hosted by team Activeloop (e.g. hub://activeloop/mnist-train).
• Underlying data format is now better! Since the new format is incompatible with the prior release, you should update to the new release using pip3 install --upgrade hub .

🧭 What's Changed

• version update (#1062) @davidbuniat
• Fixes an issue in which reporting configuration file was not being created if its parent directory didn't exist. (#1058) @zomglings
• Update PR template to new format (#1059) @benchislett
• Add back PR template from master (#1034) @benchislett
• Update htype docs (#1030) @benchislett
• Validate indexing when given, not at compute-time (#1033) @benchislett
• Update readme (#1057) @istranic
• fix meta non-persistence bug (adds test) (#1053) @mccrearyd
• Updating old pytorch warning message (#1055) @AbhinavTuli
• Changed from master to main (#1052) @Anselmoo
• [refactor] Tests/update fixtures (#1046) @mccrearyd
• NPZ replacement format (only) (#1047) @farizrahman4u
• Auto cast (#1041) @farizrahman4u
• Bring back tuple mode, this time serializable (#1028) @farizrahman4u
• Array interface for Tensor (#1042) @farizrahman4u
• Windows always uses old pytorch integration now (#1044) @AbhinavTuli
• [small] remove chunk sizes from htypes (#1037) @mccrearyd
• Small fix for Pytorch shared memory leak (#1040) @AbhinavTuli
• Fixes dataset creation bug with s3/hub cloud datasets having similar names (#1045) @AbhinavTuli
• [small] Update/2.0/hub cloud test (#1023) @mccrearyd
• Fix tensor creation bug (#1043) @farizrahman4u
• Refactor/fstrings (#1035) @dhiganthrao
• update sample compression API (#1038) @mccrearyd
• [small] Silence tensorflow logs in tests (#1029) @benchislett
• [small] update scalar test (#1022) @mccrearyd

🐛 Bug Fixes

• [small] pytorch readonly error bug fix (#1026) @mccrearyd
• [small] Fix/2.0/readonly (#1024) @mccrearyd

🔗 Dependency Updates

• Bump pillow from 7.2.0 to 8.2.0 in /requirements (#1018) @dependabot

⚙️ Who Contributes

@AbhinavTuli, @Anselmoo, @benchislett, @davidbuniat, @dependabot, @dhiganthrao, @farizrahman4u, @istranic, @mccrearyd, @tatevikh and @zomglings