Possible error when the QA retrieval_gt shape will be different (#774)

* add apply_recursive function at util

* resolve possible error at the run.py retrieval

* resolve possible error in other run.py while using retrieval_gt

* add qa evolve api docs rst

---------

Co-authored-by: jeffrey <vkefhdl1@gmail.com>

autorag/nodes/passageaugmenter/run.py

@@ -8,6 +8,7 @@
 from autorag.nodes.retrieval.run import evaluate_retrieval_node
 from autorag.schema.metricinput import MetricInput
 from autorag.strategy import measure_speed, filter_by_threshold, select_best
+from autorag.utils.util import apply_recursive, to_list
 
 logger = logging.getLogger("AutoRAG")
 
@@ -26,13 +27,7 @@ def run_passage_augmenter_node(
 		os.path.join(project_dir, "data", "qa.parquet"), engine="pyarrow"
 	)
 	retrieval_gt = qa_df["retrieval_gt"].tolist()
-	retrieval_gt = [
-		[
-			[str(uuid) for uuid in sub_array] if sub_array.size > 0 else []
-			for sub_array in inner_array
-		]
-		for inner_array in retrieval_gt
-	]
+	retrieval_gt = apply_recursive(lambda x: str(x), to_list(retrieval_gt))
 
 	results, execution_times = zip(
 		*map(

autorag/nodes/passagefilter/run.py

@@ -7,6 +7,7 @@
 from autorag.nodes.retrieval.run import evaluate_retrieval_node
 from autorag.schema.metricinput import MetricInput
 from autorag.strategy import measure_speed, filter_by_threshold, select_best
+from autorag.utils.util import to_list, apply_recursive
 
 
 def run_passage_filter_node(
@@ -37,13 +38,8 @@ def run_passage_filter_node(
 		os.path.join(project_dir, "data", "qa.parquet"), engine="pyarrow"
 	)
 	retrieval_gt = qa_df["retrieval_gt"].tolist()
-	retrieval_gt = [
-		[
-			[str(uuid) for uuid in sub_array] if sub_array.size > 0 else []
-			for sub_array in inner_array
-		]
-		for inner_array in retrieval_gt
-	]
+	retrieval_gt = apply_recursive(lambda x: str(x), to_list(retrieval_gt))
+
 	# make rows to metric_inputs
 	metric_inputs = [
 		MetricInput(retrieval_gt=ret_gt, query=query, generation_gt=gen_gt)

autorag/nodes/passagereranker/run.py

@@ -8,6 +8,7 @@
 from autorag.nodes.retrieval.run import evaluate_retrieval_node
 from autorag.schema.metricinput import MetricInput
 from autorag.strategy import measure_speed, filter_by_threshold, select_best
+from autorag.utils.util import apply_recursive, to_list
 
 logger = logging.getLogger("AutoRAG")
 
@@ -40,13 +41,8 @@ def run_passage_reranker_node(
 		os.path.join(project_dir, "data", "qa.parquet"), engine="pyarrow"
 	)
 	retrieval_gt = qa_df["retrieval_gt"].tolist()
-	retrieval_gt = [
-		[
-			[str(uuid) for uuid in sub_array] if sub_array.size > 0 else []
-			for sub_array in inner_array
-		]
-		for inner_array in retrieval_gt
-	]
+	retrieval_gt = apply_recursive(lambda x: str(x), to_list(retrieval_gt))
+
 	# make rows to metric_inputs
 	metric_inputs = [
 		MetricInput(retrieval_gt=ret_gt, query=query, generation_gt=gen_gt)

autorag/nodes/retrieval/run.py

@@ -12,7 +12,7 @@
 from autorag.schema.metricinput import MetricInput
 from autorag.strategy import measure_speed, filter_by_threshold, select_best
 from autorag.support import get_support_modules
-from autorag.utils.util import get_best_row, to_list
+from autorag.utils.util import get_best_row, to_list, apply_recursive
 
 logger = logging.getLogger("AutoRAG")
 
@@ -47,13 +47,7 @@ def run_retrieval_node(
 		os.path.join(project_dir, "data", "qa.parquet"), engine="pyarrow"
 	)
 	retrieval_gt = qa_df["retrieval_gt"].tolist()
-	retrieval_gt = [
-		[
-			[str(uuid) for uuid in sub_array] if sub_array.size > 0 else []
-			for sub_array in inner_array
-		]
-		for inner_array in retrieval_gt
-	]
+	retrieval_gt = apply_recursive(lambda x: str(x), to_list(retrieval_gt))
 	# make rows to metric_inputs
 	metric_inputs = [
 		MetricInput(retrieval_gt=ret_gt, query=query, generation_gt=gen_gt)

autorag/utils/util.py

@@ -620,3 +620,23 @@ def pop_params(func: Callable, kwargs: Dict) -> Dict:
 		if key in target_params:
 			init_params[key] = kwargs.pop(key)
 	return init_params
+
+
+def apply_recursive(func, data):
+	"""
+	Recursively apply a function to all elements in a list, tuple, set, np.ndarray, or pd.Series and return as List.
+
+	:param func: Function to apply to each element.
+	:param data: List or nested list.
+	:return: List with the function applied to each element.
+	"""
+	if (
+		isinstance(data, list)
+		or isinstance(data, tuple)
+		or isinstance(data, set)
+		or isinstance(data, np.ndarray)
+		or isinstance(data, pd.Series)
+	):
+		return [apply_recursive(func, item) for item in data]
+	else:
+		return func(data)

docs/source/api_spec/autorag.data.qa.evolve.rst

@@ -0,0 +1,37 @@
+autorag.data.qa.evolve package
+==============================
+
+Submodules
+----------
+
+autorag.data.qa.evolve.llama\_index\_query\_evolve module
+---------------------------------------------------------
+
+.. automodule:: autorag.data.qa.evolve.llama_index_query_evolve
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+autorag.data.qa.evolve.openai\_query\_evolve module
+---------------------------------------------------
+
+.. automodule:: autorag.data.qa.evolve.openai_query_evolve
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+autorag.data.qa.evolve.prompt module
+------------------------------------
+
+.. automodule:: autorag.data.qa.evolve.prompt
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Module contents
+---------------
+
+.. automodule:: autorag.data.qa.evolve
+   :members:
+   :undoc-members:
+   :show-inheritance:

tests/autorag/nodes/retrieval/test_run_retrieval_node.py

@@ -26,7 +26,7 @@
 def node_line_dir():
 	with tempfile.TemporaryDirectory(ignore_cleanup_errors=True) as test_project_dir:
 		sample_project_dir = os.path.join(resources_dir, "sample_project")
-		# copy & paste all folders and files in sample_project folder
+		# copy & paste all folders and files in the sample_project folder
 		shutil.copytree(sample_project_dir, test_project_dir, dirs_exist_ok=True)
 
 		chroma_path = os.path.join(test_project_dir, "resources", "chroma")
@@ -39,9 +39,9 @@ def node_line_dir():
 		corpus_df = pd.read_parquet(corpus_path)
 		vectordb_ingest(collection, corpus_df, MockEmbedding(1536))
 
-		test_trail_dir = os.path.join(test_project_dir, "test_trial")
-		os.makedirs(test_trail_dir)
-		node_line_dir = os.path.join(test_trail_dir, "test_node_line")
+		test_trial_dir = os.path.join(test_project_dir, "test_trial")
+		os.makedirs(test_trial_dir)
+		node_line_dir = os.path.join(test_trial_dir, "test_node_line")
 		os.makedirs(node_line_dir)
 		yield node_line_dir
 

tests/autorag/utils/test_util.py

@@ -37,6 +37,7 @@
 	get_event_loop,
 	find_key_values,
 	pop_params,
+	apply_recursive,
 )
 from tests.mock import MockLLM
 
@@ -609,3 +610,29 @@ def func_mixed(param1, param2, *args, **kwargs):
 	result = pop_params(func_mixed, kwargs)
 	assert result == expected
 	assert kwargs == {"extra_param": "extra_value"}
+
+
+def test_apply_recursive():
+	data = [1, 2, 3, 4]
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == [2, 4, 6, 8]
+
+	data = [[1, 2], [3, 4]]
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == [[2, 4], [6, 8]]
+
+	data = [[1, [2, 3]], [4, [5, 6]]]
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == [[2, [4, 6]], [8, [10, 12]]]
+
+	data = []
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == []
+
+	data = 5
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == 10
+
+	data = [(4, 5), (6, 7), [5, [6, 7]], np.array([4, 5]), pd.Series([4, 5])]
+	result = apply_recursive(lambda x: x * 2, data)
+	assert result == [[8, 10], [12, 14], [10, [12, 14]], [8, 10], [8, 10]]