NLP pipeline validation with CI tests

[fmigneault]

changes

• add JSON typings
• align JSON formatting across files
• fix nlp imports
• add GitHub CI tests workflow

[fmigneault]

@TimeaBagosiCrim
Imports are now validated and working.
However, one test seems to fail:
tests/test_MetricsClasses.py::MetricsClassesTests::test_val - AssertionError: 1.0 != 0
https://github.com/crim-ca/pavics-jupyter-images/actions/runs/7121350381/job/19390424450?pr=22

Do you know what could be the cause?

[fmigneault]

@TimeaBagosiCrim

The error seems to be introduced by this step:

pavics-jupyter-images/nlp/notebooks/nl2q_eval/MetricsClasses.py

Lines 839 to 843 in 3301b62

	for value_type in VALUE_TYPES:
	value_measures.get_value_metrics(value_type).perfect_value_match = \
	value_measures.get_value_metrics(value_type).perfect_value_match \
	/ value_measures.get_value_metrics(value_type).total_matching_attributes \
	if value_measures.get_value_metrics(value_type).total_matching_attributes > 0 else 0

Before running this loop, value_measures.get_value_metrics("numeric").perfect_value_match returns 1.0 as expected for the first annotation with "value":

pavics-jupyter-images/nlp/tests/gold_queries.json

Lines 72 to 80 in 3301b62

	{
	"text": "cloud cover lower than 10%",
	"position": [57, 83],
	"type": "property",
	"name": "cloud cover",
	"value": 10,
	"value_type": "percentage",
	"operation": "lt"
	}

The total_matching_attributes is always zero because this doesn't succeed (actual int is in value, not a numeric str) :

pavics-jupyter-images/nlp/notebooks/nl2q_eval/MetricsClasses.py

Lines 797 to 799 in 3301b62

	if 'value' in ann.keys() and isinstance(ann['value'], str):
	# print(ann['value'])
	if isnumeric(ann['value']):

This lead to resetting perfect_value_match to zero each time for "numeric" because of the loop.

Is it a problem to adjust the logic to allow int/float values as well, or will that break other code somewhere else that expects str only?