Skip to content

Commit

Permalink
More info into 0.6.0 release notes.
Browse files Browse the repository at this point in the history
  • Loading branch information
Yevgeni Litvin authored and jsgoller1 committed Feb 21, 2019
1 parent b55a6c2 commit 032e487
Showing 1 changed file with 13 additions and 3 deletions.
16 changes: 13 additions & 3 deletions docs/release-notes.rst
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,8 @@ Release notes
Release 0.6.0
=============

Thanks to our new contributers: Kim Hammar and Joshua Goller!

Breaking changes
----------------
- ``petastorm.etl.dataset_metadata.materialize_dataset()`` should be passed a filesystem factory method
Expand All @@ -15,12 +17,20 @@ Breaking changes

New features and bug fixes
--------------------------
- Added functionality for transform-on-worker thread/pool. The transform enables PyTorch users to run preprocessing
code on worker processes/threads. It enables Tensorflow users to parallelize Python preprocessing code on
a process pool, as part of the training/evaluation graph. Users now specify a ``transform_spec`` when calling
``make_reader()`` or ``make_batch_reader()``.
- Added ``hdfs_driver`` argument to the following functions: ``get_schema_from_dataset_url``, ``FilesystemResolver``,
``generate_petastorm_metadata``, ``build_rowgroup_index``, ``RowGroupLoader``, ``dataset_as_rdd`` and ``copy_dataset``
- the Docker container in ``/docker`` has been made into a workspace container aimed to support development on MacOS.
- New `hello_world` examples added for using non-Petastorm datasets.
- Added functionality for transform-on-worker thread/pool. Users now specify a ``transform_spec`` when calling ``make_reader()``
or ``make_batch_reader()``
- Fixed a bug that caused all columns of a dataset to be read when ``schema_fields=NGram(...)`` was used.
- Allow for unicode strings to be passed as regex filters in Unischema when selecting which columns to read.
- Fixed a bug that caused all columns of a dataset to be read when ``schema_fields=NGram(...)`` was used.
- Fixed type of an argument passed to a predicate when the predicate is defined on a numeric partition field
- Support regular unicode strings as expressions as a value of make_reader's schema_fields argument.
- Emit a warning when opening a Petastorm-created dataset using make_batch_reader (``make_batch_reader`` currently
does not support Petastorm specific types, such as tensors).

Release 0.5.1
=============
Expand Down

0 comments on commit 032e487

Please sign in to comment.