Sophia Koehler
commited on
Commit
·
d661944
1
Parent(s):
3f7f963
fix
Browse files- app.py +494 -8
- requirements.txt +2 -0
app.py
CHANGED
@@ -1,3 +1,488 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
from typing import TypedDict
|
3 |
|
@@ -8,24 +493,25 @@ class Hit(TypedDict):
|
|
8 |
|
9 |
demo: Optional[gr.Interface] = None # Assign your gradio demo to this variable
|
10 |
return_type = List[Hit]
|
|
|
11 |
## YOUR_CODE_STARTS_HERE
|
12 |
def search_sciq(query: str) -> List[Hit]:
|
13 |
results = bm25_retriever.retrieve(query)
|
14 |
-
|
15 |
-
# Format the output to match the List[Hit] structure
|
16 |
-
hits = []
|
17 |
for cid, score in results.items():
|
18 |
index = bm25_retriever.index.cid2docid[cid]
|
19 |
text = bm25_retriever.index.doc_texts[index]
|
20 |
-
|
21 |
|
22 |
-
return
|
23 |
|
24 |
-
# Set up the Gradio interface
|
25 |
demo = gr.Interface(
|
26 |
fn=search_sciq,
|
27 |
-
inputs=
|
28 |
-
outputs=
|
29 |
description="BM25 Search Engine Demo on SciQ Dataset"
|
30 |
)
|
|
|
31 |
demo.launch()
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
"""Kopie von HW1 (more instructed).ipynb
|
3 |
+
|
4 |
+
Automatically generated by Colab.
|
5 |
+
|
6 |
+
Original file is located at
|
7 |
+
https://colab.research.google.com/drive/1BrX2Zy737ji-Lbb2evMV2P-WfzvTniHj
|
8 |
+
"""
|
9 |
+
|
10 |
+
!pip install git+https://github.com/kwang2049/nlp4web-codebase.git
|
11 |
+
!git clone https://github.com/kwang2049/nlp4web-codebase.git # You can always check the content of this simple codebase at any time
|
12 |
+
!pip install gradio # we also need this additionally for this homework
|
13 |
+
|
14 |
+
"""## Pre-requisite code
|
15 |
+
|
16 |
+
The code within this section will be used in the tasks. Please do not change these code lines.
|
17 |
+
|
18 |
+
### SciQ loading and counting
|
19 |
+
"""
|
20 |
+
|
21 |
+
from dataclasses import dataclass
|
22 |
+
import pickle
|
23 |
+
import os
|
24 |
+
from typing import Iterable, Callable, List, Dict, Optional, Type, TypeVar
|
25 |
+
from nlp4web_codebase.ir.data_loaders.dm import Document
|
26 |
+
from collections import Counter
|
27 |
+
import tqdm
|
28 |
+
import re
|
29 |
+
import nltk
|
30 |
+
nltk.download("stopwords", quiet=True)
|
31 |
+
from nltk.corpus import stopwords as nltk_stopwords
|
32 |
+
|
33 |
+
LANGUAGE = "english"
|
34 |
+
word_splitter = re.compile(r"(?u)\b\w\w+\b").findall
|
35 |
+
stopwords = set(nltk_stopwords.words(LANGUAGE))
|
36 |
+
|
37 |
+
|
38 |
+
def word_splitting(text: str) -> List[str]:
|
39 |
+
return word_splitter(text.lower())
|
40 |
+
|
41 |
+
def lemmatization(words: List[str]) -> List[str]:
|
42 |
+
return words # We ignore lemmatization here for simplicity
|
43 |
+
|
44 |
+
def simple_tokenize(text: str) -> List[str]:
|
45 |
+
words = word_splitting(text)
|
46 |
+
tokenized = list(filter(lambda w: w not in stopwords, words))
|
47 |
+
tokenized = lemmatization(tokenized)
|
48 |
+
return tokenized
|
49 |
+
|
50 |
+
T = TypeVar("T", bound="InvertedIndex")
|
51 |
+
|
52 |
+
@dataclass
|
53 |
+
class PostingList:
|
54 |
+
term: str # The term
|
55 |
+
docid_postings: List[int] # docid_postings[i] means the docid (int) of the i-th associated posting
|
56 |
+
tweight_postings: List[float] # tweight_postings[i] means the term weight (float) of the i-th associated posting
|
57 |
+
|
58 |
+
|
59 |
+
@dataclass
|
60 |
+
class InvertedIndex:
|
61 |
+
posting_lists: List[PostingList] # docid -> posting_list
|
62 |
+
vocab: Dict[str, int]
|
63 |
+
cid2docid: Dict[str, int] # collection_id -> docid
|
64 |
+
collection_ids: List[str] # docid -> collection_id
|
65 |
+
doc_texts: Optional[List[str]] = None # docid -> document text
|
66 |
+
|
67 |
+
def save(self, output_dir: str) -> None:
|
68 |
+
os.makedirs(output_dir, exist_ok=True)
|
69 |
+
with open(os.path.join(output_dir, "index.pkl"), "wb") as f:
|
70 |
+
pickle.dump(self, f)
|
71 |
+
|
72 |
+
@classmethod
|
73 |
+
def from_saved(cls: Type[T], saved_dir: str) -> T:
|
74 |
+
index = cls(
|
75 |
+
posting_lists=[], vocab={}, cid2docid={}, collection_ids=[], doc_texts=None
|
76 |
+
)
|
77 |
+
with open(os.path.join(saved_dir, "index.pkl"), "rb") as f:
|
78 |
+
index = pickle.load(f)
|
79 |
+
return index
|
80 |
+
|
81 |
+
|
82 |
+
# The output of the counting function:
|
83 |
+
@dataclass
|
84 |
+
class Counting:
|
85 |
+
posting_lists: List[PostingList]
|
86 |
+
vocab: Dict[str, int]
|
87 |
+
cid2docid: Dict[str, int]
|
88 |
+
collection_ids: List[str]
|
89 |
+
dfs: List[int] # tid -> df
|
90 |
+
dls: List[int] # docid -> doc length
|
91 |
+
avgdl: float
|
92 |
+
nterms: int
|
93 |
+
doc_texts: Optional[List[str]] = None
|
94 |
+
|
95 |
+
def run_counting(
|
96 |
+
documents: Iterable[Document],
|
97 |
+
tokenize_fn: Callable[[str], List[str]] = simple_tokenize,
|
98 |
+
store_raw: bool = True, # store the document text in doc_texts
|
99 |
+
ndocs: Optional[int] = None,
|
100 |
+
show_progress_bar: bool = True,
|
101 |
+
) -> Counting:
|
102 |
+
"""Counting TFs, DFs, doc_lengths, etc."""
|
103 |
+
posting_lists: List[PostingList] = []
|
104 |
+
vocab: Dict[str, int] = {}
|
105 |
+
cid2docid: Dict[str, int] = {}
|
106 |
+
collection_ids: List[str] = []
|
107 |
+
dfs: List[int] = [] # tid -> df
|
108 |
+
dls: List[int] = [] # docid -> doc length
|
109 |
+
nterms: int = 0
|
110 |
+
doc_texts: Optional[List[str]] = []
|
111 |
+
for doc in tqdm.tqdm(
|
112 |
+
documents,
|
113 |
+
desc="Counting",
|
114 |
+
total=ndocs,
|
115 |
+
disable=not show_progress_bar,
|
116 |
+
):
|
117 |
+
if doc.collection_id in cid2docid:
|
118 |
+
continue
|
119 |
+
collection_ids.append(doc.collection_id)
|
120 |
+
docid = cid2docid.setdefault(doc.collection_id, len(cid2docid))
|
121 |
+
toks = tokenize_fn(doc.text)
|
122 |
+
tok2tf = Counter(toks)
|
123 |
+
dls.append(sum(tok2tf.values()))
|
124 |
+
for tok, tf in tok2tf.items():
|
125 |
+
nterms += tf
|
126 |
+
tid = vocab.get(tok, None)
|
127 |
+
if tid is None:
|
128 |
+
posting_lists.append(
|
129 |
+
PostingList(term=tok, docid_postings=[], tweight_postings=[])
|
130 |
+
)
|
131 |
+
tid = vocab.setdefault(tok, len(vocab))
|
132 |
+
posting_lists[tid].docid_postings.append(docid)
|
133 |
+
posting_lists[tid].tweight_postings.append(tf)
|
134 |
+
if tid < len(dfs):
|
135 |
+
dfs[tid] += 1
|
136 |
+
else:
|
137 |
+
dfs.append(0)
|
138 |
+
if store_raw:
|
139 |
+
doc_texts.append(doc.text)
|
140 |
+
else:
|
141 |
+
doc_texts = None
|
142 |
+
return Counting(
|
143 |
+
posting_lists=posting_lists,
|
144 |
+
vocab=vocab,
|
145 |
+
cid2docid=cid2docid,
|
146 |
+
collection_ids=collection_ids,
|
147 |
+
dfs=dfs,
|
148 |
+
dls=dls,
|
149 |
+
avgdl=sum(dls) / len(dls),
|
150 |
+
nterms=nterms,
|
151 |
+
doc_texts=doc_texts,
|
152 |
+
)
|
153 |
+
|
154 |
+
from nlp4web_codebase.ir.data_loaders.sciq import load_sciq
|
155 |
+
sciq = load_sciq()
|
156 |
+
counting = run_counting(documents=iter(sciq.corpus), ndocs=len(sciq.corpus))
|
157 |
+
|
158 |
+
"""### BM25 Index"""
|
159 |
+
|
160 |
+
from __future__ import annotations
|
161 |
+
from dataclasses import asdict, dataclass
|
162 |
+
import math
|
163 |
+
import os
|
164 |
+
from typing import Iterable, List, Optional, Type
|
165 |
+
import tqdm
|
166 |
+
from nlp4web_codebase.ir.data_loaders.dm import Document
|
167 |
+
|
168 |
+
|
169 |
+
@dataclass
|
170 |
+
class BM25Index(InvertedIndex):
|
171 |
+
|
172 |
+
@staticmethod
|
173 |
+
def tokenize(text: str) -> List[str]:
|
174 |
+
return simple_tokenize(text)
|
175 |
+
|
176 |
+
@staticmethod
|
177 |
+
def cache_term_weights(
|
178 |
+
posting_lists: List[PostingList],
|
179 |
+
total_docs: int,
|
180 |
+
avgdl: float,
|
181 |
+
dfs: List[int],
|
182 |
+
dls: List[int],
|
183 |
+
k1: float,
|
184 |
+
b: float,
|
185 |
+
) -> None:
|
186 |
+
"""Compute term weights and caching"""
|
187 |
+
|
188 |
+
N = total_docs
|
189 |
+
for tid, posting_list in enumerate(
|
190 |
+
tqdm.tqdm(posting_lists, desc="Regularizing TFs")
|
191 |
+
):
|
192 |
+
idf = BM25Index.calc_idf(df=dfs[tid], N=N)
|
193 |
+
for i in range(len(posting_list.docid_postings)):
|
194 |
+
docid = posting_list.docid_postings[i]
|
195 |
+
tf = posting_list.tweight_postings[i]
|
196 |
+
dl = dls[docid]
|
197 |
+
regularized_tf = BM25Index.calc_regularized_tf(
|
198 |
+
tf=tf, dl=dl, avgdl=avgdl, k1=k1, b=b
|
199 |
+
)
|
200 |
+
posting_list.tweight_postings[i] = regularized_tf * idf
|
201 |
+
|
202 |
+
@staticmethod
|
203 |
+
def calc_regularized_tf(
|
204 |
+
tf: int, dl: float, avgdl: float, k1: float, b: float
|
205 |
+
) -> float:
|
206 |
+
return tf / (tf + k1 * (1 - b + b * dl / avgdl))
|
207 |
+
|
208 |
+
@staticmethod
|
209 |
+
def calc_idf(df: int, N: int):
|
210 |
+
return math.log(1 + (N - df + 0.5) / (df + 0.5))
|
211 |
+
|
212 |
+
@classmethod
|
213 |
+
def build_from_documents(
|
214 |
+
cls: Type[BM25Index],
|
215 |
+
documents: Iterable[Document],
|
216 |
+
store_raw: bool = True,
|
217 |
+
output_dir: Optional[str] = None,
|
218 |
+
ndocs: Optional[int] = None,
|
219 |
+
show_progress_bar: bool = True,
|
220 |
+
k1: float = 0.9,
|
221 |
+
b: float = 0.4,
|
222 |
+
) -> BM25Index:
|
223 |
+
# Counting TFs, DFs, doc_lengths, etc.:
|
224 |
+
counting = run_counting(
|
225 |
+
documents=documents,
|
226 |
+
tokenize_fn=BM25Index.tokenize,
|
227 |
+
store_raw=store_raw,
|
228 |
+
ndocs=ndocs,
|
229 |
+
show_progress_bar=show_progress_bar,
|
230 |
+
)
|
231 |
+
|
232 |
+
# Compute term weights and caching:
|
233 |
+
posting_lists = counting.posting_lists
|
234 |
+
total_docs = len(counting.cid2docid)
|
235 |
+
BM25Index.cache_term_weights(
|
236 |
+
posting_lists=posting_lists,
|
237 |
+
total_docs=total_docs,
|
238 |
+
avgdl=counting.avgdl,
|
239 |
+
dfs=counting.dfs,
|
240 |
+
dls=counting.dls,
|
241 |
+
k1=k1,
|
242 |
+
b=b,
|
243 |
+
)
|
244 |
+
|
245 |
+
# Assembly and save:
|
246 |
+
index = BM25Index(
|
247 |
+
posting_lists=posting_lists,
|
248 |
+
vocab=counting.vocab,
|
249 |
+
cid2docid=counting.cid2docid,
|
250 |
+
collection_ids=counting.collection_ids,
|
251 |
+
doc_texts=counting.doc_texts,
|
252 |
+
)
|
253 |
+
return index
|
254 |
+
|
255 |
+
bm25_index = BM25Index.build_from_documents(
|
256 |
+
documents=iter(sciq.corpus),
|
257 |
+
ndocs=12160,
|
258 |
+
show_progress_bar=True,
|
259 |
+
)
|
260 |
+
bm25_index.save("output/bm25_index")
|
261 |
+
!ls
|
262 |
+
|
263 |
+
"""### BM25 Retriever"""
|
264 |
+
|
265 |
+
from nlp4web_codebase.ir.models import BaseRetriever
|
266 |
+
from typing import Type
|
267 |
+
from abc import abstractmethod
|
268 |
+
|
269 |
+
|
270 |
+
class BaseInvertedIndexRetriever(BaseRetriever):
|
271 |
+
|
272 |
+
@property
|
273 |
+
@abstractmethod
|
274 |
+
def index_class(self) -> Type[InvertedIndex]:
|
275 |
+
pass
|
276 |
+
|
277 |
+
def __init__(self, index_dir: str) -> None:
|
278 |
+
self.index = self.index_class.from_saved(index_dir)
|
279 |
+
|
280 |
+
def get_term_weights(self, query: str, cid: str) -> Dict[str, float]:
|
281 |
+
toks = self.index.tokenize(query)
|
282 |
+
target_docid = self.index.cid2docid[cid]
|
283 |
+
term_weights = {}
|
284 |
+
for tok in toks:
|
285 |
+
if tok not in self.index.vocab:
|
286 |
+
continue
|
287 |
+
tid = self.index.vocab[tok]
|
288 |
+
posting_list = self.index.posting_lists[tid]
|
289 |
+
for docid, tweight in zip(
|
290 |
+
posting_list.docid_postings, posting_list.tweight_postings
|
291 |
+
):
|
292 |
+
if docid == target_docid:
|
293 |
+
term_weights[tok] = tweight
|
294 |
+
break
|
295 |
+
return term_weights
|
296 |
+
|
297 |
+
def score(self, query: str, cid: str) -> float:
|
298 |
+
return sum(self.get_term_weights(query=query, cid=cid).values())
|
299 |
+
|
300 |
+
def retrieve(self, query: str, topk: int = 10) -> Dict[str, float]:
|
301 |
+
toks = self.index.tokenize(query)
|
302 |
+
docid2score: Dict[int, float] = {}
|
303 |
+
for tok in toks:
|
304 |
+
if tok not in self.index.vocab:
|
305 |
+
continue
|
306 |
+
tid = self.index.vocab[tok]
|
307 |
+
posting_list = self.index.posting_lists[tid]
|
308 |
+
for docid, tweight in zip(
|
309 |
+
posting_list.docid_postings, posting_list.tweight_postings
|
310 |
+
):
|
311 |
+
docid2score.setdefault(docid, 0)
|
312 |
+
docid2score[docid] += tweight
|
313 |
+
docid2score = dict(
|
314 |
+
sorted(docid2score.items(), key=lambda pair: pair[1], reverse=True)[:topk]
|
315 |
+
)
|
316 |
+
return {
|
317 |
+
self.index.collection_ids[docid]: score
|
318 |
+
for docid, score in docid2score.items()
|
319 |
+
}
|
320 |
+
|
321 |
+
|
322 |
+
class BM25Retriever(BaseInvertedIndexRetriever):
|
323 |
+
|
324 |
+
@property
|
325 |
+
def index_class(self) -> Type[BM25Index]:
|
326 |
+
return BM25Index
|
327 |
+
|
328 |
+
bm25_retriever = BM25Retriever(index_dir="output/bm25_index")
|
329 |
+
bm25_retriever.retrieve("What type of diseases occur when the immune system attacks normal body cells?")
|
330 |
+
|
331 |
+
"""# TASK1: tune b and k1 (4 points)
|
332 |
+
|
333 |
+
Tune b and k1 on the **dev** split of SciQ using the metric MAP@10. The evaluation function (`evalaute_map`) is provided. Record the values in `plots_k1` and `plots_b`. Do it in a greedy manner: as the influence from b is larger, please first tune b (with k1 fixed to the default value 0.9) and use the best value of b to further tune k1.
|
334 |
+
|
335 |
+
$${\displaystyle {\text{score}}(D,Q)=\sum _{i=1}^{n}{\text{IDF}}(q_{i})\cdot {\frac {f(q_{i},D)\cdot (k_{1}+1)}{f(q_{i},D)+k_{1}\cdot \left(1-b+b\cdot {\frac {|D|}{\text{avgdl}}}\right)}}}$$
|
336 |
+
"""
|
337 |
+
|
338 |
+
from nlp4web_codebase.ir.data_loaders import Split
|
339 |
+
import pytrec_eval
|
340 |
+
|
341 |
+
|
342 |
+
def evaluate_map(rankings: Dict[str, Dict[str, float]], split=Split.dev) -> float:
|
343 |
+
metric = "map_cut_10"
|
344 |
+
qrels = sciq.get_qrels_dict(split)
|
345 |
+
evaluator = pytrec_eval.RelevanceEvaluator(sciq.get_qrels_dict(split), (metric,))
|
346 |
+
qps = evaluator.evaluate(rankings)
|
347 |
+
return float(np.mean([qp[metric] for qp in qps.values()]))
|
348 |
+
|
349 |
+
"""Example of using the pre-requisite code:"""
|
350 |
+
|
351 |
+
# Loading dataset:
|
352 |
+
from nlp4web_codebase.ir.data_loaders.sciq import load_sciq
|
353 |
+
sciq = load_sciq()
|
354 |
+
counting = run_counting(documents=iter(sciq.corpus), ndocs=len(sciq.corpus))
|
355 |
+
|
356 |
+
# Building BM25 index and save:
|
357 |
+
bm25_index = BM25Index.build_from_documents(
|
358 |
+
documents=iter(sciq.corpus),
|
359 |
+
ndocs=12160,
|
360 |
+
show_progress_bar=True
|
361 |
+
)
|
362 |
+
bm25_index.save("output/bm25_index")
|
363 |
+
|
364 |
+
# Loading index and use BM25 retriever to retrieve:
|
365 |
+
bm25_retriever = BM25Retriever(index_dir="output/bm25_index")
|
366 |
+
print(bm25_retriever.retrieve("What type of diseases occur when the immune system attacks normal body cells?")) # the ranking
|
367 |
+
|
368 |
+
plots_b: Dict[str, List[float]] = {
|
369 |
+
"X": [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
|
370 |
+
"Y": []
|
371 |
+
}
|
372 |
+
plots_k1: Dict[str, List[float]] = {
|
373 |
+
"X": [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
|
374 |
+
"Y": []
|
375 |
+
}
|
376 |
+
|
377 |
+
## YOUR_CODE_STARTS_HERE
|
378 |
+
class MyBMIndex(BM25Index):
|
379 |
+
|
380 |
+
@staticmethod
|
381 |
+
def calc_regularized_tf(
|
382 |
+
tf: int, dl: float, avgdl: float, k1: float, b: float
|
383 |
+
) -> float:
|
384 |
+
return tf * (k1 + 1) / (tf + k1 * (1 - b + b * (dl / avgdl)**1.5))
|
385 |
+
|
386 |
+
@staticmethod
|
387 |
+
def calc_idf(df: int, N: int):
|
388 |
+
return math.log((N + 1) / (df + 0.5)) + 1
|
389 |
+
import numpy as np
|
390 |
+
# Two steps should be involved:
|
391 |
+
# Step 1. Fix k1 value to the default one 0.9,
|
392 |
+
# go through all the candidate b values (0, 0.1, ..., 1.0),
|
393 |
+
# and record in plots_b["Y"] the corresponding performances obtained via evaluate_map;
|
394 |
+
# Step 2. Fix b to the best one in step 1. and do the same for k1.
|
395 |
+
|
396 |
+
# Hint (on using the pre-requisite code):
|
397 |
+
# - One can use the loaded sciq dataset directly (loaded in the pre-requisite code);
|
398 |
+
# - One can build bm25_index with `BM25Index.build_from_documents`;
|
399 |
+
# - One can use BM25Retriever to load the index and perform retrieval on the dev queries
|
400 |
+
# (dev queries can be obtained via sciq.get_split_queries(Split.dev))
|
401 |
+
|
402 |
+
counting = run_counting(documents=iter(sciq.corpus), ndocs=len(sciq.corpus))
|
403 |
+
|
404 |
+
def get_ranking(k1, b, counting) -> Dict[str, Dict[str, float]]:
|
405 |
+
# Building BM25 index and save:
|
406 |
+
bm25_index = MyBMIndex.build_from_documents(
|
407 |
+
documents=iter(sciq.corpus),
|
408 |
+
ndocs=12160,
|
409 |
+
show_progress_bar=True,
|
410 |
+
k1=k1,
|
411 |
+
b=b
|
412 |
+
)
|
413 |
+
bm25_index.save("output/bm25_index")
|
414 |
+
|
415 |
+
# Loading index and use BM25 retriever to retrieve:
|
416 |
+
bm25_retriever = BM25Retriever(index_dir="output/bm25_index")
|
417 |
+
query_terms = sciq.get_split_queries(split= Split.dev)
|
418 |
+
rankings = {}
|
419 |
+
for query in query_terms:
|
420 |
+
ranking = bm25_retriever.retrieve(query=query.text)
|
421 |
+
rankings[query.query_id] = ranking
|
422 |
+
return rankings
|
423 |
+
for b in plots_b["X"]:
|
424 |
+
ranking = get_ranking(0.9, b, counting)
|
425 |
+
plots_b["Y"].append(evaluate_map(rankings=ranking))
|
426 |
+
|
427 |
+
max_b = np.max(plots_b["Y"])
|
428 |
+
for k1 in plots_k1["X"]:
|
429 |
+
ranking = get_ranking(k1, max_b, counting)
|
430 |
+
plots_k1["Y"].append(evaluate_map(rankings=ranking))
|
431 |
+
## YOU_CODE_ENDS_HERE
|
432 |
+
|
433 |
+
## TEST_CASES (should be close to 0.8135637188208616 and 0.7512916099773244)
|
434 |
+
print(plots_k1["Y"][9])
|
435 |
+
print(plots_b["Y"][1])
|
436 |
+
|
437 |
+
## RESULT_CHECKING_POINT
|
438 |
+
print(plots_k1)
|
439 |
+
print(plots_b)
|
440 |
+
|
441 |
+
from matplotlib import pyplot as plt
|
442 |
+
plt.plot(plots_b["X"], plots_b["Y"], label="b")
|
443 |
+
plt.plot(plots_k1["X"], plots_k1["Y"], label="k1")
|
444 |
+
plt.ylabel("MAP")
|
445 |
+
plt.legend()
|
446 |
+
plt.grid()
|
447 |
+
plt.show()
|
448 |
+
|
449 |
+
"""Let's check the effectiveness gain on test after this tuning on dev"""
|
450 |
+
|
451 |
+
default_map = 0.7849
|
452 |
+
best_b = plots_b["X"][np.argmax(plots_b["Y"])]
|
453 |
+
best_k1 = plots_k1["X"][np.argmax(plots_k1["Y"])]
|
454 |
+
bm25_index = BM25Index.build_from_documents(
|
455 |
+
documents=iter(sciq.corpus),
|
456 |
+
ndocs=12160,
|
457 |
+
show_progress_bar=True,
|
458 |
+
k1=best_k1,
|
459 |
+
b=best_b
|
460 |
+
)
|
461 |
+
bm25_index.save("output/bm25_index")
|
462 |
+
bm25_retriever = BM25Retriever(index_dir="output/bm25_index")
|
463 |
+
rankings = {}
|
464 |
+
for query in sciq.get_split_queries(Split.test): # note this is now on test
|
465 |
+
ranking = bm25_retriever.retrieve(query=query.text)
|
466 |
+
rankings[query.query_id] = ranking
|
467 |
+
optimized_map = evaluate_map(rankings, split=Split.test) # note this is now on test
|
468 |
+
print(default_map, optimized_map)
|
469 |
+
|
470 |
+
"""# TASK3: a search-engine demo based on Huggingface space (4 points)
|
471 |
+
|
472 |
+
## TASK3.1: create the gradio app (2 point)
|
473 |
+
|
474 |
+
Create a gradio app to demo the BM25 search engine index on SciQ. The app should have a single input variable for the query (of type `str`) and a single output variable for the returned ranking (of type `List[Hit]` in the code below). Please use the BM25 system with default k1 and b values.
|
475 |
+
|
476 |
+
Hint: it should use a "search" function of signature:
|
477 |
+
|
478 |
+
```python
|
479 |
+
def search(query: str) -> List[Hit]:
|
480 |
+
...
|
481 |
+
```
|
482 |
+
"""
|
483 |
+
|
484 |
+
!pip install gradio
|
485 |
+
|
486 |
import gradio as gr
|
487 |
from typing import TypedDict
|
488 |
|
|
|
493 |
|
494 |
demo: Optional[gr.Interface] = None # Assign your gradio demo to this variable
|
495 |
return_type = List[Hit]
|
496 |
+
|
497 |
## YOUR_CODE_STARTS_HERE
|
498 |
def search_sciq(query: str) -> List[Hit]:
|
499 |
results = bm25_retriever.retrieve(query)
|
500 |
+
hitlist = []
|
|
|
|
|
501 |
for cid, score in results.items():
|
502 |
index = bm25_retriever.index.cid2docid[cid]
|
503 |
text = bm25_retriever.index.doc_texts[index]
|
504 |
+
hitlist.append(Hit(cid=cid, score=score, text=text))
|
505 |
|
506 |
+
return hitlist
|
507 |
|
|
|
508 |
demo = gr.Interface(
|
509 |
fn=search_sciq,
|
510 |
+
inputs="textbox",
|
511 |
+
outputs="textbox",
|
512 |
description="BM25 Search Engine Demo on SciQ Dataset"
|
513 |
)
|
514 |
+
## YOUR_CODE_ENDS_HERE
|
515 |
demo.launch()
|
516 |
+
|
517 |
+
|
requirements.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
gradio
|
2 |
+
git+https://github.com/kwang2049/nlp4web-codebase.git
|