IBM / unitxt / 13941206858

81%

push

github

Changed API of Key Value Extraction task to use Dict and not List[Tuple] (NON BACKWARD COMPATIBLE CHANGE) (#1675)

* Moved key value extraction task  and metrics to use Dict[str,str] to compare and not List[Tuple[str,str]]

Also created a dedicated metric.

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Updated metric to better address in case keys are used in prediction and not references

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Added ability to customize inner metric used to compare each entity in KeyValueExtraction

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fixed unitest and bug with handling nones

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fixed KeyValueExtraction prepare

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Added example of multiple metrics

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fix catalog

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fix some bugs in inference engine tests

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Fix some bugs in inference engine tests

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Updated key value extraction metric names

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Updated key value extraction metric names

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Updated documentation string

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fixed unit test.

* Updated to use metric as artifact and not string

* Fix bug in name conversion in rits

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Add engine id

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Improved output message when using inference cache

Also fixed issue when all data was in the cache and an empty list was passed to _infer.

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fixed bug due to indentation change

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* fix

Signed-off-by: elronbandel <elronbandel@gmail.com>

* fix

Signed-off-by: elronbandel <elronbandel@gmail.com>

* Removed warning of legacy name.

Sign... (continued)

1566 of 1932 branches covered (81.06%)

Branch coverage included in aggregate %.

9794 of 12131 relevant lines covered (80.74%)

0.81 hits per line