Major Features and Improvements
- Saved tuner results in pandas
recordsformatted JSON. - TFX Transform now supports
tf.SequenceExamplenatively. The native path can be activated by providingTensorRepresentations in the Schema. - TFX Transform now supports reading raw and materializing transformed data in
Apache Parquet format. - ExampleDiff outputs statistics on the matching process, and optional counts
of paired feature values.
Breaking Changes
- N/A
For Pipeline Authors
- N/A
For Component Authors
- N/A
Deprecations
- N/A
Bug Fixes and Other Changes
- Type hint on BaseComponent.inputs and BaseComponent.outputs corrected to be
Channel subclasses. - Added
input_optionalparameter toChannelParameter. This allows
component authors to declare that even if a channel isoptional, if it is
provided during pipeline definition time, then it must have resolved inputs
during run time. - Allow latest
apache-airflow2.x versions. - Moved
tflite-supportrelated dependencies from[examples]to a separate
[tflite-support]extra. - Output artifacts from multiple invocations of the same component are given
unique names, avoiding duplication errors, especially in the
InteractiveContext.
Dependency Updates
| Package Name | Version Constraints | Previously (in v1.9.0)
| Comments |
|---|---|---|---|
google-api-core
| <2
| N/A | Added to help pip dependency resolution. google-api-core was already a transitive dependency. |
apache-beam[gcp]
| >=2.40,<3
| >=2.38,<3
| Synced release train |
attrs
| >=19.3.0,<22
| >=19.3.0,<21
| Allow more recent versions |
pyarrow
| >=6,<7
| >=1,<6
| Synced release train |
ml-metadata
| ~=1.10.0
| ~=1.9.0
| Synced release train |
struct2tensor
| ~=0.41.0
| ~=0.40.0
| Synced release train |
tensorflow-data-validation
| ~=1.10.0
| ~=1.9.0
| Synced release train |
tensorflow-model-analysis
| ~=0.41.0
| ~=0.40.0
| Synced release train |
tensorflow-transform
| ~=1.10.1
| ~=1.9.0
| Synced release train |
tfx-bsl
| ~=1.10.1
| ~=1.9.0
| Synced release train |
Documentation Updates
- N/A