Spaces:
Runtime error
Runtime error
Commit
·
aad2fb9
1
Parent(s):
45c093d
Update app.py
Browse files
app.py
CHANGED
@@ -15,118 +15,139 @@ st.set_page_config(page_title="PSC Runtime",
|
|
15 |
page_icon='🌸', layout="centered")
|
16 |
|
17 |
|
18 |
-
import torch
|
19 |
-
fn = "dl19-gpt-3.5.pt"
|
20 |
-
object = torch.load(fn)
|
21 |
|
22 |
-
|
23 |
-
outputs = object[2]
|
24 |
-
query2outputs = {}
|
25 |
-
for output in outputs:
|
26 |
-
all_queries = {x['query'] for x in output}
|
27 |
-
assert len(all_queries) == 1
|
28 |
-
query = list(all_queries)[0]
|
29 |
-
query2outputs[query] = [x['hits'] for x in output]
|
30 |
-
|
31 |
-
|
32 |
-
search_query = st.selectbox(
|
33 |
"",
|
34 |
-
|
35 |
index=None,
|
36 |
-
placeholder="Choose a
|
37 |
)
|
38 |
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
for doc in result:
|
46 |
-
if doc["docid"] not in docid2id:
|
47 |
-
id = len(docid2id)
|
48 |
-
docid2id[doc["docid"]] = id
|
49 |
-
id2doc[id] = doc
|
50 |
-
print([doc["docid"] for doc in result])
|
51 |
-
print([docid2id[doc["docid"]] for doc in result])
|
52 |
-
preferences.append([docid2id[doc["docid"]] for doc in result])
|
53 |
-
|
54 |
-
# = {v: k for k, v in docid2id.items()}
|
55 |
-
return np.array(preferences), id2doc
|
56 |
-
|
57 |
-
|
58 |
-
def load_qrels(name):
|
59 |
-
import ir_datasets
|
60 |
-
if name == "dl19":
|
61 |
-
ds_name = "msmarco-passage/trec-dl-2019/judged"
|
62 |
-
elif name == "dl20":
|
63 |
-
ds_name = "msmarco-passage/trec-dl-2020/judged"
|
64 |
-
else:
|
65 |
-
raise ValueError(name)
|
66 |
-
|
67 |
-
dataset = ir_datasets.load(ds_name)
|
68 |
-
qrels = defaultdict(dict)
|
69 |
-
for qrel in dataset.qrels_iter():
|
70 |
-
qrels[qrel.query_id][qrel.doc_id] = qrel.relevance
|
71 |
-
return qrels
|
72 |
-
|
73 |
-
|
74 |
-
def aggregate(list_of_hits):
|
75 |
-
import numpy as np
|
76 |
-
from permsc import KemenyOptimalAggregator, sum_kendall_tau, ranks_from_preferences
|
77 |
-
from permsc import BordaRankAggregator
|
78 |
-
|
79 |
-
preferences, id2doc = preferences_from_hits(list_of_hits)
|
80 |
-
y_optimal = KemenyOptimalAggregator().aggregate(preferences)
|
81 |
-
# y_optimal = BordaRankAggregator().aggregate(preferences)
|
82 |
-
|
83 |
-
return [id2doc[id] for id in y_optimal]
|
84 |
-
|
85 |
-
|
86 |
-
def write_ranking(search_results):
|
87 |
-
# st.write(
|
88 |
-
# f'<p align=\"right\" style=\"color:grey;\"> Before aggregation for query [{search_query}] ms</p>', unsafe_allow_html=True)
|
89 |
-
|
90 |
-
qid = {result["qid"] for result in search_results}
|
91 |
-
assert len(qid) == 1
|
92 |
-
qid = list(qid)[0]
|
93 |
|
94 |
-
|
95 |
-
result_id = result["docid"]
|
96 |
-
contents = result["content"]
|
97 |
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
else:
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
page_icon='🌸', layout="centered")
|
16 |
|
17 |
|
|
|
|
|
|
|
18 |
|
19 |
+
name = st.selectbox(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
"",
|
21 |
+
["dl19", "dl20"]
|
22 |
index=None,
|
23 |
+
placeholder="Choose a dataset..."
|
24 |
)
|
25 |
|
26 |
+
model_name = st.selectbox(
|
27 |
+
"",
|
28 |
+
["gpt-3.5", "gpt-4"]
|
29 |
+
index=None,
|
30 |
+
placeholder="Choose a model..."
|
31 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
+
# "dl19"
|
|
|
|
|
34 |
|
35 |
+
if name and model_name:
|
36 |
+
|
37 |
+
import torch
|
38 |
+
# fn = f"dl19-gpt-3.5.pt"
|
39 |
+
fn = f"{name}-{model_name}.pt"
|
40 |
+
object = torch.load(fn)
|
41 |
+
|
42 |
+
|
43 |
+
outputs = object[2]
|
44 |
+
query2outputs = {}
|
45 |
+
for output in outputs:
|
46 |
+
all_queries = {x['query'] for x in output}
|
47 |
+
assert len(all_queries) == 1
|
48 |
+
query = list(all_queries)[0]
|
49 |
+
query2outputs[query] = [x['hits'] for x in output]
|
50 |
+
|
51 |
+
|
52 |
+
search_query = st.selectbox(
|
53 |
+
"",
|
54 |
+
sorted(query2outputs),
|
55 |
+
index=None,
|
56 |
+
placeholder="Choose a query from the list..."
|
57 |
+
)
|
58 |
+
|
59 |
+
def preferences_from_hits(list_of_hits):
|
60 |
+
docid2id = {}
|
61 |
+
id2doc = {}
|
62 |
+
preferences = []
|
63 |
+
|
64 |
+
for result in list_of_hits:
|
65 |
+
for doc in result:
|
66 |
+
if doc["docid"] not in docid2id:
|
67 |
+
id = len(docid2id)
|
68 |
+
docid2id[doc["docid"]] = id
|
69 |
+
id2doc[id] = doc
|
70 |
+
print([doc["docid"] for doc in result])
|
71 |
+
print([docid2id[doc["docid"]] for doc in result])
|
72 |
+
preferences.append([docid2id[doc["docid"]] for doc in result])
|
73 |
+
|
74 |
+
# = {v: k for k, v in docid2id.items()}
|
75 |
+
return np.array(preferences), id2doc
|
76 |
+
|
77 |
+
|
78 |
+
def load_qrels(name):
|
79 |
+
import ir_datasets
|
80 |
+
if name == "dl19":
|
81 |
+
ds_name = "msmarco-passage/trec-dl-2019/judged"
|
82 |
+
elif name == "dl20":
|
83 |
+
ds_name = "msmarco-passage/trec-dl-2020/judged"
|
84 |
else:
|
85 |
+
raise ValueError(name)
|
86 |
+
|
87 |
+
dataset = ir_datasets.load(ds_name)
|
88 |
+
qrels = defaultdict(dict)
|
89 |
+
for qrel in dataset.qrels_iter():
|
90 |
+
qrels[qrel.query_id][qrel.doc_id] = qrel.relevance
|
91 |
+
return qrels
|
92 |
+
|
93 |
+
|
94 |
+
def aggregate(list_of_hits):
|
95 |
+
import numpy as np
|
96 |
+
from permsc import KemenyOptimalAggregator, sum_kendall_tau, ranks_from_preferences
|
97 |
+
from permsc import BordaRankAggregator
|
98 |
+
|
99 |
+
preferences, id2doc = preferences_from_hits(list_of_hits)
|
100 |
+
y_optimal = KemenyOptimalAggregator().aggregate(preferences)
|
101 |
+
# y_optimal = BordaRankAggregator().aggregate(preferences)
|
102 |
+
|
103 |
+
return [id2doc[id] for id in y_optimal]
|
104 |
+
|
105 |
+
|
106 |
+
def write_ranking(search_results):
|
107 |
+
# st.write(
|
108 |
+
# f'<p align=\"right\" style=\"color:grey;\"> Before aggregation for query [{search_query}] ms</p>', unsafe_allow_html=True)
|
109 |
+
|
110 |
+
qid = {result["qid"] for result in search_results}
|
111 |
+
assert len(qid) == 1
|
112 |
+
qid = list(qid)[0]
|
113 |
+
|
114 |
+
for i, result in enumerate(search_results):
|
115 |
+
result_id = result["docid"]
|
116 |
+
contents = result["content"]
|
117 |
+
|
118 |
+
label = qrels[str(qid)].get(str(result_id), 0)
|
119 |
+
if label == 3:
|
120 |
+
style = "style=\"color:blue;\""
|
121 |
+
elif label == 2:
|
122 |
+
style = "style=\"color:green;\""
|
123 |
+
elif label == 1:
|
124 |
+
style = "style=\"color:red;\""
|
125 |
+
else:
|
126 |
+
style = "style=\"color:grey;\""
|
127 |
+
|
128 |
+
print(qid, result_id, label, style)
|
129 |
+
# output = f'<div class="row"> <b>Rank</b>: {i+1} | <b>Document ID</b>: {result_id} | <b>Score</b>:{result_score:.2f}</div>'
|
130 |
+
output = f'<div class="row" {style}> <b>Rank</b>: {i+1} | <b>Document ID</b>: {result_id}'
|
131 |
+
|
132 |
+
try:
|
133 |
+
st.write(output, unsafe_allow_html=True)
|
134 |
+
st.write(
|
135 |
+
f'<div class="row" {style}>{contents}</div>', unsafe_allow_html=True)
|
136 |
+
|
137 |
+
except:
|
138 |
+
pass
|
139 |
+
st.write('---')
|
140 |
+
|
141 |
+
|
142 |
+
aggregated_ranking = aggregate(query2outputs[search_query])
|
143 |
+
qrels = load_qrels(name)
|
144 |
+
col1, col2 = st.columns([5, 5])
|
145 |
+
|
146 |
+
if search_query:
|
147 |
+
with col1:
|
148 |
+
if search_query or button_clicked:
|
149 |
+
write_ranking(search_results=query2outputs[search_query][0])
|
150 |
+
|
151 |
+
with col2:
|
152 |
+
if search_query or button_clicked:
|
153 |
+
write_ranking(search_results=aggregated_ranking)
|