def generate_basic_graph_with_mapping(
VISJS=False) -> (GraphDataStruct, dict):
mapping = {}
# Create a graphe structure
if VISJS:
tmp_meta = Metadata(Source.VISJS)
else:
tmp_meta = Metadata(Source.DBDUMP)
tmp_graph = GraphDataStruct(tmp_meta)
# For each cluster, fetch all pictures and store it
for cluster_id in range(0, 2):
tmp_graph.add_cluster(
Cluster(label="", tmp_id=cluster_id, image=""))
for id in range(0, 3):
pic_id = str(cluster_id) + "_" + str(id) + "OLD"
pic_image = str(cluster_id) + "_" + str(id) + "IMAGE"
# Prepare mapping
mapping[pic_image] = str(cluster_id) + "_" + str(id) + "NEW"
# Label = picture score, here
tmp_graph.add_node(
Node(label="picture name +" + pic_id,
tmp_id=pic_id,
image=pic_image))
tmp_graph.add_edge(Edge(_from=cluster_id, _to=pic_id))
return tmp_graph, mapping
def get_storage_graph(self) -> GraphDataStruct:
"""
# TODO : Move to API ?
Export the current state of the database as a graph datastructure. This represents the storage graph of the server.
:return: The storage graph of the server, as is in the database
"""
# Create a graphe structure
tmp_meta = Metadata(Source.DBDUMP)
tmp_graph = GraphDataStruct(tmp_meta)
# Get all clusters
cluster_list = self.get_cluster_list()
# For each cluster, fetch all pictures and store it
for cluster_id in cluster_list:
tmp_graph.add_cluster(Cluster(label="", tmp_id=cluster_id, image=""))
picture_list = self.get_pictures_of_cluster(cluster_id, with_score=True)
self.logger.info(f"Picture list : {picture_list}")
for picture in picture_list:
# Label = picture score, here
tmp_graph.add_node(Node(label=picture[1], tmp_id=picture[0], image=""))
tmp_graph.add_edge(Edge(_from=cluster_id, _to=picture[0]))
return tmp_graph
def get_basic_graph():
graph_1 = graph_datastructure.GraphDataStruct(Metadata(Source.DBDUMP))
graph_1.add_cluster(
Cluster(label="cluster_1", tmp_id="c1", image="image_c1"))
graph_1.add_cluster(
Cluster(label="cluster_2", tmp_id="c2", image="image_c2"))
graph_1.add_cluster(
Cluster(label="cluster_3", tmp_id="c3", image="image_c3"))
graph_1.add_node(Node(label="node_1", tmp_id="n1", image="image_n1"))
graph_1.add_node(Node(label="node_2", tmp_id="n2", image="image_n2"))
graph_1.add_node(Node(label="node_3", tmp_id="n3", image="image_n3"))
graph_1.add_node(Node(label="node_4", tmp_id="n4", image="image_n4"))
graph_1.add_edge(Edge(_from="n1", _to="c1"))
graph_1.add_edge(Edge(_from="n4", _to="c1"))
graph_1.add_edge(Edge(_from="n2", _to="c2"))
graph_1.add_edge(Edge(_from="n3", _to="c3"))
return graph_1
def generate_basic_graph() -> GraphDataStruct:
# Create a graphe structure
tmp_meta = Metadata(Source.DBDUMP)
tmp_graph = GraphDataStruct(tmp_meta)
# For each cluster, fetch all pictures and store it
for cluster_id in range(0, 2):
tmp_graph.add_cluster(
Cluster(label="", tmp_id=cluster_id, image=""))
for id in range(0, 3):
pic_id = str(cluster_id) + "_" + str(id)
# Label = picture score, here
tmp_graph.add_node(
Node(label="picture name +" + pic_id,
tmp_id=pic_id,
image=""))
tmp_graph.add_edge(Edge(_from=cluster_id, _to=pic_id))
return tmp_graph
def test_compute_score_for_one_threshold(self):
# Graph example. Please check documentation for more information
cal_conf = Default_calibrator_conf()
quality_evaluator = similarity_graph_quality_evaluator.similarity_graph_quality_evaluator(
cal_conf)
requests_results = [
# 1 to 2 and 3
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.1,
"image_id": "2"
}, {
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.6,
"image_id": "3"
}],
"request_id":
"1",
"status":
"matches_found",
"request_time":
0
},
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.3,
"image_id": "6"
}],
"request_id":
"2",
"status":
"matches_found",
"request_time":
0
},
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.2,
"image_id": "1"
}],
"request_id":
"3",
"status":
"matches_found",
"request_time":
0
},
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.4,
"image_id": "2"
}],
"request_id":
"4",
"status":
"matches_found",
"request_time":
0
},
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.6,
"image_id": "4"
}],
"request_id":
"5",
"status":
"matches_found",
"request_time":
0
},
{
"list_pictures": [{
"cluster_id": "XXX",
"decision": "YES",
"distance": 0.7,
"image_id": "2"
}],
"request_id":
"6",
"status":
"matches_found",
"request_time":
0
}
]
# Create the reference graph
graph_data_struct = GraphDataStruct(Metadata(Source.DBDUMP))
graph_data_struct.add_cluster(Cluster(label="A", tmp_id="A",
image="A"))
graph_data_struct.add_cluster(Cluster(label="B", tmp_id="B",
image="B"))
graph_data_struct.add_cluster(Cluster(label="C", tmp_id="C",
image="C"))
graph_data_struct.add_node(Node(label="1", tmp_id="1", image="1"))
graph_data_struct.add_node(Node(label="2", tmp_id="2", image="2"))
graph_data_struct.add_node(Node(label="3", tmp_id="3", image="3"))
graph_data_struct.add_node(Node(label="4", tmp_id="4", image="4"))
graph_data_struct.add_node(Node(label="5", tmp_id="5", image="5"))
graph_data_struct.add_node(Node(label="6", tmp_id="6", image="6"))
graph_data_struct.add_edge(Edge(_from="1", _to="A"))
graph_data_struct.add_edge(Edge(_from="2", _to="A"))
graph_data_struct.add_edge(Edge(_from="3", _to="A"))
graph_data_struct.add_edge(Edge(_from="4", _to="B"))
graph_data_struct.add_edge(Edge(_from="5", _to="B"))
graph_data_struct.add_edge(Edge(_from="6", _to="C"))
pprint.pprint(requests_results)
pprint.pprint(graph_data_struct.export_as_dict())
quality_evaluator.cal_conf.NB_TO_CHECK = 1
dist_threshold = 0
stats_datastruct = quality_evaluator.compute_score_for_one_threshold(
requests_results, graph_data_struct, dist_threshold)
print(stats_datastruct)
self.assertEqual(stats_datastruct.P, 3)
self.assertEqual(stats_datastruct.N, 3)
self.assertAlmostEqual(stats_datastruct.TPR, 0.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.TNR, 1.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FPR, 0.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FNR, 1.0, delta=0.05)
dist_threshold = 0.5
stats_datastruct = quality_evaluator.compute_score_for_one_threshold(
requests_results, graph_data_struct, dist_threshold)
print(stats_datastruct)
self.assertEqual(stats_datastruct.P, 3)
self.assertEqual(stats_datastruct.N, 3)
self.assertAlmostEqual(stats_datastruct.TPR, 0.66, delta=0.05)
self.assertAlmostEqual(stats_datastruct.TNR, 0.33, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FPR, 0.66, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FNR, 0.33, delta=0.05)
dist_threshold = 1
stats_datastruct = quality_evaluator.compute_score_for_one_threshold(
requests_results, graph_data_struct, dist_threshold)
print(stats_datastruct)
self.assertEqual(stats_datastruct.P, 3)
self.assertEqual(stats_datastruct.N, 3)
self.assertAlmostEqual(stats_datastruct.TPR, 1.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.TNR, 0.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FPR, 1.0, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FNR, 0.0, delta=0.05)
quality_evaluator.cal_conf.NB_TO_CHECK = 3
dist_threshold = 0.5
stats_datastruct = quality_evaluator.compute_score_for_one_threshold(
requests_results, graph_data_struct, dist_threshold)
print(stats_datastruct)
self.assertEqual(stats_datastruct.P, 4)
self.assertEqual(stats_datastruct.N, 3)
self.assertAlmostEqual(stats_datastruct.TPR, 0.5, delta=0.05)
self.assertAlmostEqual(stats_datastruct.TNR, 0.33, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FPR, 0.66, delta=0.05)
self.assertAlmostEqual(stats_datastruct.FNR, 0.5, delta=0.05)
def results_list_to_graph(requests_list, nb_match: int = 1, yes_maybe_no_mode: bool = False) -> GraphDataStruct:
"""
Construct a graph from results list of requests on the database
Hypothesis : All database pictures are requested pictures
Edges are colored : from green to red depending on the match index (Best is green)
:param requests_list: a List of results extracted from server
:param nb_match: Number of matches per pictures to add to the graph (1=first level match/best match, 2 = 2 best match per picture, etc.)
:return: A graph datastructure
"""
logger = logging.getLogger(__name__)
# logger.debug(f"Received request_list : {pformat(requests_list)}")
graph = GraphDataStruct(meta=Metadata(source=Source.DBDUMP))
# Color Managemement
# FF0000 = red
# 00FF00 = green
short_color_list = ["#00FF00", "#887700", "#CC3300", "#FF0000"]
color_list = ["#00FF00", "#11EE00", "#22DD00", "#33CC00", "#44BB00", "#55AA00",
"#669900", "#778800", "#887700", "#996600", "#AA5500", "#BB4400",
"#CC3300", "#DD2200", "#EE1100", "#FF0000"]
if nb_match < 4:
# We only have 4 colors if we don't want that much matches.
# This way, first match is green, second orange, third red, etc.
color_list = short_color_list
# TODO : Print all YES match
# For each request
for curr_req_1 in requests_list:
# logger.debug(f"Curent request : {pformat(curr_req_1)}")
req_id = curr_req_1.get("request_id", None)
# Requested picture => add a node
graph.add_node(Node(label=req_id, tmp_id=req_id, image=req_id))
# For each request
for curr_req_2 in requests_list:
# We remove the picture "itself" from the matches
tmp_clean_matches = get_clear_matches(curr_req_2)
req_id = curr_req_2.get("request_id", None)
# Add edge for each best pictures
for i in range(min(nb_match, len(tmp_clean_matches))):
dist = round(tmp_clean_matches[i].get("distance", None), 4)
deci = tmp_clean_matches[i].get("decision", "UNKNOWN")
dest_id = tmp_clean_matches[i].get("image_id", None)
if dist is None:
logger.error(f"Problem with request, no distance: {pformat(curr_req_2)}")
continue
if dest_id is None:
logger.error(f"Problem with request, no match's image id: {pformat(curr_req_2)}")
continue
if yes_maybe_no_mode:
# set threshold depending on Yes/Maybe/No in VisJS
# By creatin a fictive distance, depending on the decision
fictive_dist = scoring_datastrutures.DecisionTypes.get_fictive_dist(deci)
# Add a fictive edge
graph.add_edge(Edge(_from=req_id,
_to=dest_id,
color={"color": color_list[i % len(color_list)]},
label=deci + ":" + str(dist),
value=fictive_dist))
else:
graph.add_edge(Edge(_from=req_id,
_to=dest_id,
color={"color": color_list[i % len(color_list)]},
label=deci + ":" + str(dist),
value=dist))
return graph