def extract_cases(self, collection):
collection_name = collection[
"collection_name"] if "collection_name" in collection else None
(data_type, dimension,
metric_type) = parser.parse_ann_collection_name(collection_name)
# hdf5_source_file: The path of the source data file saved on the NAS
hdf5_source_file = collection["source_file"]
index_types = collection["index_types"]
index_params = collection["index_params"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
guarantee_timestamp = collection[
"guarantee_timestamp"] if "guarantee_timestamp" in collection else None
search_params = collection["search_params"]
vector_type = utils.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
dataset = utils.get_dataset(hdf5_source_file)
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
filters = collection["filters"] if "filters" in collection else []
filter_query = []
# Convert list data into a set of dictionary data
search_params = utils.generate_combinations(search_params)
index_params = utils.generate_combinations(index_params)
cases = list()
case_metrics = list()
self.init_metric(self.name, collection_info, {}, search_info=None)
# true_ids: The data set used to verify the results returned by query
true_ids = np.array(dataset["neighbors"])
for index_type in index_types:
for index_param in index_params:
index_info = {
"index_type": index_type,
"index_param": index_param
}
for search_param in search_params:
if not filters:
filters.append(None)
for filter in filters:
filter_param = []
if isinstance(filter, dict) and "range" in filter:
filter_query.append(eval(filter["range"]))
filter_param.append(filter["range"])
if isinstance(filter, dict) and "term" in filter:
filter_query.append(eval(filter["term"]))
filter_param.append(filter["term"])
for nq in nqs:
query_vectors = utils.normalize(
metric_type, np.array(dataset["test"][:nq]))
for top_k in top_ks:
search_info = {
"topk":
top_k,
"query":
query_vectors,
"metric_type":
utils.metric_type_trans(metric_type),
"params":
search_param
}
# TODO: only update search_info
case_metric = copy.deepcopy(self.metric)
# set metric type as case
case_metric.set_case_metric_type()
case_metric.index = index_info
case_metric.search = {
"nq": nq,
"topk": top_k,
"search_param": search_param,
"filter": filter_param,
"guarantee_timestamp": guarantee_timestamp
}
vector_query = {
"vector": {
index_field_name: search_info
}
}
case = {
"collection_name": collection_name,
"dataset": dataset,
"index_field_name": index_field_name,
"dimension": dimension,
"data_type": data_type,
"metric_type": metric_type,
"vector_type": vector_type,
"index_type": index_type,
"index_param": index_param,
"filter_query": filter_query,
"vector_query": vector_query,
"true_ids": true_ids,
"guarantee_timestamp": guarantee_timestamp
}
# Obtain the parameters of the use case to be tested
cases.append(case)
case_metrics.append(case_metric)
return cases, case_metrics
def extract_cases(self, collection):
collection_name = collection[
"collection_name"] if "collection_name" in collection else None
(data_type, dimension,
metric_type) = parser.parse_ann_collection_name(collection_name)
hdf5_source_file = collection["source_file"]
index_types = collection["index_types"]
index_params = collection["index_params"]
top_ks = collection["top_ks"]
nqs = collection["nqs"]
search_params = collection["search_params"]
vector_type = utils.get_vector_type(data_type)
index_field_name = utils.get_default_field_name(vector_type)
dataset = utils.get_dataset(hdf5_source_file)
collection_info = {
"dimension": dimension,
"metric_type": metric_type,
"dataset_name": collection_name
}
filters = collection["filters"] if "filters" in collection else []
filter_query = []
search_params = utils.generate_combinations(search_params)
index_params = utils.generate_combinations(index_params)
cases = list()
case_metrics = list()
self.init_metric(self.name, collection_info, {}, search_info=None)
true_ids = np.array(dataset["neighbors"])
for index_type in index_types:
for index_param in index_params:
index_info = {
"index_type": index_type,
"index_param": index_param
}
for search_param in search_params:
if not filters:
filters.append(None)
for filter in filters:
filter_param = []
if isinstance(filter, dict) and "range" in filter:
filter_query.append(eval(filter["range"]))
filter_param.append(filter["range"])
if isinstance(filter, dict) and "term" in filter:
filter_query.append(eval(filter["term"]))
filter_param.append(filter["term"])
for nq in nqs:
query_vectors = utils.normalize(
metric_type, np.array(dataset["test"][:nq]))
for top_k in top_ks:
search_info = {
"topk":
top_k,
"query":
query_vectors,
"metric_type":
utils.metric_type_trans(metric_type),
"params":
search_param
}
# TODO: only update search_info
case_metric = copy.deepcopy(self.metric)
case_metric.index = index_info
case_metric.search = {
"nq": nq,
"topk": top_k,
"search_param": search_param,
"filter": filter_param
}
vector_query = {
"vector": {
index_field_name: search_info
}
}
case = {
"collection_name": collection_name,
"dataset": dataset,
"index_field_name": index_field_name,
"dimension": dimension,
"data_type": data_type,
"metric_type": metric_type,
"vector_type": vector_type,
"index_type": index_type,
"index_param": index_param,
"filter_query": filter_query,
"vector_query": vector_query,
"true_ids": true_ids
}
cases.append(case)
case_metrics.append(case_metric)
return cases, case_metrics