def test_eval_pipeline(document_store: BaseDocumentStore, reader, retriever):
# add eval data (SQUAD format)
document_store.add_eval_data(
filename="samples/squad/tiny.json",
doc_index="haystack_test_eval_document",
label_index="haystack_test_feedback",
)
labels = document_store.get_all_labels_aggregated(
index="haystack_test_feedback")
q_to_l_dict = {l.question: {"retriever": l, "reader": l} for l in labels}
eval_retriever = EvalRetriever()
eval_reader = EvalReader()
assert document_store.get_document_count(
index="haystack_test_eval_document") == 2
p = Pipeline()
p.add_node(component=retriever, name="ESRetriever", inputs=["Query"])
p.add_node(component=eval_retriever,
name="EvalRetriever",
inputs=["ESRetriever"])
p.add_node(component=reader, name="QAReader", inputs=["EvalRetriever"])
p.add_node(component=eval_reader, name="EvalReader", inputs=["QAReader"])
for q, l in q_to_l_dict.items():
res = p.run(
query=q,
top_k_retriever=10,
labels=l,
top_k_reader=10,
index="haystack_test_eval_document",
)
assert eval_retriever.recall == 1.0
assert round(eval_reader.top_k_f1, 4) == 0.8333
assert eval_reader.top_k_em == 0.5
def test_eval_finder(document_store: BaseDocumentStore, reader, retriever):
finder = Finder(reader=reader, retriever=retriever)
# add eval data (SQUAD format)
document_store.add_eval_data(
filename="samples/squad/tiny.json",
doc_index="haystack_test_eval_document",
label_index="haystack_test_feedback",
)
assert document_store.get_document_count(index="haystack_test_eval_document") == 2
# eval finder
results = finder.eval(
label_index="haystack_test_feedback", doc_index="haystack_test_eval_document", top_k_retriever=1, top_k_reader=5
)
assert results["retriever_recall"] == 1.0
assert results["retriever_map"] == 1.0
assert abs(results["reader_topk_f1"] - 0.66666) < 0.001
assert abs(results["reader_topk_em"] - 0.5) < 0.001
assert abs(results["reader_topk_accuracy"] - 1) < 0.001
assert results["reader_top1_f1"] <= results["reader_topk_f1"]
assert results["reader_top1_em"] <= results["reader_topk_em"]
assert results["reader_top1_accuracy"] <= results["reader_topk_accuracy"]
# batch eval finder
results_batch = finder.eval_batch(
label_index="haystack_test_feedback", doc_index="haystack_test_eval_document", top_k_retriever=1, top_k_reader=5
)
assert results_batch["retriever_recall"] == 1.0
assert results_batch["retriever_map"] == 1.0
assert results_batch["reader_top1_f1"] == results["reader_top1_f1"]
assert results_batch["reader_top1_em"] == results["reader_top1_em"]
assert results_batch["reader_topk_accuracy"] == results["reader_topk_accuracy"]
def test_eval_elastic_retriever(document_store: BaseDocumentStore, open_domain, retriever):
# add eval data (SQUAD format)
document_store.delete_all_documents(index="test_eval_document")
document_store.delete_all_documents(index="test_feedback")
document_store.add_eval_data(filename="samples/squad/tiny.json", doc_index="test_eval_document", label_index="test_feedback")
assert document_store.get_document_count(index="test_eval_document") == 2
# eval retriever
results = retriever.eval(top_k=1, label_index="test_feedback", doc_index="test_eval_document", open_domain=open_domain)
assert results["recall"] == 1.0
assert results["map"] == 1.0
# clean up
document_store.delete_all_documents(index="test_eval_document")
document_store.delete_all_documents(index="test_feedback")
def test_eval_reader(reader, document_store: BaseDocumentStore):
# add eval data (SQUAD format)
document_store.delete_all_documents(index="test_eval_document")
document_store.delete_all_documents(index="test_feedback")
document_store.add_eval_data(filename="samples/squad/tiny.json", doc_index="test_eval_document", label_index="test_feedback")
assert document_store.get_document_count(index="test_eval_document") == 2
# eval reader
reader_eval_results = reader.eval(document_store=document_store, label_index="test_feedback",
doc_index="test_eval_document", device="cpu")
assert reader_eval_results["f1"] > 0.65
assert reader_eval_results["f1"] < 0.67
assert reader_eval_results["EM"] == 0.5
assert reader_eval_results["top_n_accuracy"] == 1.0
# clean up
document_store.delete_all_documents(index="test_eval_document")
document_store.delete_all_documents(index="test_feedback")
def test_eval_reader(reader, document_store: BaseDocumentStore):
# add eval data (SQUAD format)
document_store.add_eval_data(
filename="samples/squad/tiny.json",
doc_index="haystack_test_eval_document",
label_index="haystack_test_feedback",
)
assert document_store.get_document_count(index="haystack_test_eval_document") == 2
# eval reader
reader_eval_results = reader.eval(
document_store=document_store,
label_index="haystack_test_feedback",
doc_index="haystack_test_eval_document",
device="cpu",
)
assert reader_eval_results["f1"] > 66.65
assert reader_eval_results["f1"] < 66.67
assert reader_eval_results["EM"] == 50
assert reader_eval_results["top_n_accuracy"] == 100.0
def test_eval_pipeline(document_store: BaseDocumentStore, reader, retriever):
# add eval data (SQUAD format)
document_store.add_eval_data(
filename="samples/squad/tiny.json",
doc_index="haystack_test_eval_document",
label_index="haystack_test_feedback",
)
labels = document_store.get_all_labels_aggregated(index="haystack_test_feedback")
eval_retriever = EvalDocuments()
eval_reader = EvalAnswers(sas_model="sentence-transformers/paraphrase-MiniLM-L3-v2",debug=True)
eval_reader_cross = EvalAnswers(sas_model="cross-encoder/stsb-TinyBERT-L-4",debug=True)
eval_reader_vanila = EvalAnswers()
assert document_store.get_document_count(index="haystack_test_eval_document") == 2
p = Pipeline()
p.add_node(component=retriever, name="ESRetriever", inputs=["Query"])
p.add_node(component=eval_retriever, name="EvalDocuments", inputs=["ESRetriever"])
p.add_node(component=reader, name="QAReader", inputs=["EvalDocuments"])
p.add_node(component=eval_reader, name="EvalAnswers", inputs=["QAReader"])
p.add_node(component=eval_reader_cross, name="EvalAnswers_cross", inputs=["QAReader"])
p.add_node(component=eval_reader_vanila, name="EvalAnswers_vanilla", inputs=["QAReader"])
for l in labels:
res = p.run(
query=l.question,
top_k_retriever=10,
labels=l,
top_k_reader=10,
index="haystack_test_eval_document",
)
assert eval_retriever.recall == 1.0
assert round(eval_reader.top_k_f1, 4) == 0.8333
assert eval_reader.top_k_em == 0.5
assert round(eval_reader.top_k_sas, 3) == 0.800
assert round(eval_reader_cross.top_k_sas, 3) == 0.671
assert eval_reader.top_k_em == eval_reader_vanila.top_k_em