fadliaulawi
commited on
Commit
•
63bec36
1
Parent(s):
8503206
Remove unnecessary function
Browse files- process.py +4 -166
process.py
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
from datetime import datetime
|
2 |
from dotenv import load_dotenv
|
3 |
from img2table.document import Image
|
4 |
from langchain.chains.combine_documents.map_reduce import MapReduceDocumentsChain
|
@@ -9,7 +8,7 @@ from langchain.prompts import PromptTemplate
|
|
9 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
10 |
from langchain_openai import ChatOpenAI
|
11 |
from pdf2image import convert_from_path
|
12 |
-
from prompt import
|
13 |
from table_detector import detection_transform, device, model, ocr, outputs_to_objects
|
14 |
|
15 |
import io
|
@@ -17,8 +16,6 @@ import json
|
|
17 |
import os
|
18 |
import pandas as pd
|
19 |
import re
|
20 |
-
import requests
|
21 |
-
import time
|
22 |
import torch
|
23 |
|
24 |
load_dotenv()
|
@@ -31,7 +28,7 @@ prompts = {
|
|
31 |
|
32 |
class Process():
|
33 |
|
34 |
-
def __init__(self, llm
|
35 |
|
36 |
if llm.startswith('gpt'):
|
37 |
self.llm = ChatOpenAI(temperature=0, model_name=llm)
|
@@ -40,13 +37,6 @@ class Process():
|
|
40 |
else:
|
41 |
self.llm = ChatOpenAI(temperature=0, model_name=llm, api_key=os.environ['PERPLEXITY_API_KEY'], base_url="https://api.perplexity.ai")
|
42 |
|
43 |
-
if llm_val.startswith('gpt'):
|
44 |
-
self.llm_val = ChatOpenAI(temperature=0, model_name=llm_val)
|
45 |
-
elif llm_val.startswith('gemini'):
|
46 |
-
self.llm_val = ChatGoogleGenerativeAI(temperature=0, model=llm_val)
|
47 |
-
else:
|
48 |
-
self.llm_val = ChatOpenAI(temperature=0, model_name=llm_val, api_key=os.environ['PERPLEXITY_API_KEY'], base_url="https://api.perplexity.ai")
|
49 |
-
|
50 |
def get_entity(self, data):
|
51 |
|
52 |
chunks, types = data
|
@@ -97,9 +87,7 @@ class Process():
|
|
97 |
|
98 |
def get_table(self, path):
|
99 |
|
100 |
-
start_time = datetime.now()
|
101 |
images = convert_from_path(path)
|
102 |
-
print('PDF to Image', round((datetime.now().timestamp() - start_time.timestamp()) / 60, 2), "minutes")
|
103 |
tables = []
|
104 |
|
105 |
# Loop pages
|
@@ -124,7 +112,6 @@ class Process():
|
|
124 |
print(detected_tables[idx])
|
125 |
tables.append(cropped_table)
|
126 |
|
127 |
-
print('Detect table from image', round((datetime.now().timestamp() - start_time.timestamp()) / 60, 2), "minutes")
|
128 |
genes = []
|
129 |
snps = []
|
130 |
diseases = []
|
@@ -147,29 +134,10 @@ class Process():
|
|
147 |
for extracted_table in extracted_tables[1:]:
|
148 |
df_table = pd.concat([df_table, extracted_table.df]).reset_index(drop=True)
|
149 |
|
150 |
-
df_table
|
151 |
-
|
152 |
-
# Identify multiple rows (in dataframe) as one row (in image)
|
153 |
-
rows = []
|
154 |
-
indexes = []
|
155 |
-
for i in df_table.index:
|
156 |
-
if not df_table.loc[i].isna().any():
|
157 |
-
if len(indexes) > 0:
|
158 |
-
rows.append(indexes)
|
159 |
-
indexes = []
|
160 |
-
indexes.append(i)
|
161 |
-
rows.append(indexes)
|
162 |
-
|
163 |
-
df_table_cleaned = pd.DataFrame(columns=df_table.columns)
|
164 |
-
for row in rows:
|
165 |
-
row_str = df_table.loc[row[0]]
|
166 |
-
for idx in row[1:]:
|
167 |
-
row_str += ' ' + df_table.loc[idx].fillna('')
|
168 |
-
row_str = row_str.str.strip()
|
169 |
-
df_table_cleaned.loc[len(df_table_cleaned)] = row_str
|
170 |
|
171 |
# Ask LLM with JSON data
|
172 |
-
json_table =
|
173 |
str_json_table = json.dumps(json.loads(json_table), indent=2)
|
174 |
|
175 |
result = self.llm.invoke(prompt_table.format(str_json_table)).content
|
@@ -191,135 +159,5 @@ class Process():
|
|
191 |
snps.append(snp)
|
192 |
diseases.append(res_disease)
|
193 |
|
194 |
-
print('OCR table to extract', round((datetime.now().timestamp() - start_time.timestamp()) / 60, 2), "minutes")
|
195 |
print(genes, snps, diseases)
|
196 |
-
|
197 |
return genes, snps, diseases
|
198 |
-
|
199 |
-
def validate(self, df):
|
200 |
-
|
201 |
-
df = df.fillna('')
|
202 |
-
df['Genes'] = df['Genes'].str.replace(' ', '').str.upper()
|
203 |
-
df['SNPs'] = df['SNPs'].str.lower()
|
204 |
-
|
205 |
-
# Check if there is two gene names
|
206 |
-
sym = [',', '/', '|']
|
207 |
-
for i in df.index:
|
208 |
-
gene = df.loc[i, 'Genes']
|
209 |
-
for s in sym:
|
210 |
-
if s in gene:
|
211 |
-
genes = gene.split(s)
|
212 |
-
df.loc[i + 0.5] = df.loc[i]
|
213 |
-
df = df.sort_index().reset_index(drop=True)
|
214 |
-
df.loc[i, 'Genes'], df.loc[i + 1, 'Genes'] = genes[0], s.join(genes[1:])
|
215 |
-
break
|
216 |
-
|
217 |
-
# Check if there is SNPs without 'rs'
|
218 |
-
for i in df.index:
|
219 |
-
safe = True
|
220 |
-
snp = df.loc[i, 'SNPs']
|
221 |
-
snp = snp.replace('l', '1')
|
222 |
-
if re.fullmatch('rs(\d)+|', snp):
|
223 |
-
pass
|
224 |
-
elif re.fullmatch('ts(\d)+', snp):
|
225 |
-
snp = 'r' + snp[1:]
|
226 |
-
elif re.fullmatch('s(\d)+', snp):
|
227 |
-
snp = 'r' + snp
|
228 |
-
elif re.fullmatch('(\d)+', snp):
|
229 |
-
snp = 'rs' + snp
|
230 |
-
else:
|
231 |
-
safe = False
|
232 |
-
df = df.drop(i)
|
233 |
-
|
234 |
-
if safe:
|
235 |
-
df.loc[i, 'SNPs'] = snp
|
236 |
-
|
237 |
-
df.reset_index(drop=True, inplace=True)
|
238 |
-
df_clean = df.copy()
|
239 |
-
|
240 |
-
# # Validate genes and SNPs with APIs
|
241 |
-
def permutate(word):
|
242 |
-
|
243 |
-
if len(word) == 0:
|
244 |
-
return ['']
|
245 |
-
|
246 |
-
change = []
|
247 |
-
res = permutate(word[1:])
|
248 |
-
|
249 |
-
if word[0] in mistakes:
|
250 |
-
change = [mistakes[word[0]] + r for r in res]
|
251 |
-
|
252 |
-
return [word[0] + r for r in res] + change
|
253 |
-
|
254 |
-
def call(url):
|
255 |
-
|
256 |
-
while True:
|
257 |
-
try:
|
258 |
-
res = requests.get(url)
|
259 |
-
time.sleep(1)
|
260 |
-
break
|
261 |
-
except Exception as e:
|
262 |
-
print(e)
|
263 |
-
|
264 |
-
return res
|
265 |
-
|
266 |
-
mistakes = {'I': '1', 'O': '0'} # Common mistakes need to be maintained
|
267 |
-
dbsnp = {}
|
268 |
-
|
269 |
-
for i in df.index:
|
270 |
-
snp = df.loc[i, 'SNPs']
|
271 |
-
gene = df.loc[i, 'Genes']
|
272 |
-
|
273 |
-
if snp not in dbsnp:
|
274 |
-
res = call(f'https://www.ebi.ac.uk/gwas/rest/api/singleNucleotidePolymorphisms/{snp}/')
|
275 |
-
try:
|
276 |
-
res = res.json()
|
277 |
-
dbsnp[snp] = [r['gene']['geneName'] for r in res['genomicContexts']]
|
278 |
-
except:
|
279 |
-
dbsnp[snp] = []
|
280 |
-
|
281 |
-
res = call(f'https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi?db=snp&retmode=json&id={snp[2:]}').json()['result'][snp[2:]]
|
282 |
-
if 'error' not in res:
|
283 |
-
dbsnp[snp].extend([r['name'] for r in res['genes']])
|
284 |
-
|
285 |
-
dbsnp[snp] = list(set(dbsnp[snp]))
|
286 |
-
|
287 |
-
if gene not in dbsnp[snp]:
|
288 |
-
for other in permutate(gene):
|
289 |
-
if other in dbsnp[snp]:
|
290 |
-
df.loc[i, 'Genes'] = other
|
291 |
-
print(f'{gene} corrected to {other}')
|
292 |
-
break
|
293 |
-
else:
|
294 |
-
df = df.drop(i)
|
295 |
-
|
296 |
-
# df.reset_index(drop=True, inplace=True)
|
297 |
-
df_no_llm = df.copy()
|
298 |
-
|
299 |
-
# Validate genes and diseases with LLM (for each 50 rows)
|
300 |
-
idx = 0
|
301 |
-
results = []
|
302 |
-
|
303 |
-
while True:
|
304 |
-
json_table = df[['Genes', 'SNPs', 'Diseases']][idx:idx+50].to_json(orient='records')
|
305 |
-
str_json_table = json.dumps(json.loads(json_table), indent=2)
|
306 |
-
|
307 |
-
result = self.llm_val.invoke(input=prompt_validation.format(str_json_table)).content
|
308 |
-
print('val', idx)
|
309 |
-
print(result)
|
310 |
-
|
311 |
-
result = result[result.find('['):result.rfind(']')+1]
|
312 |
-
try:
|
313 |
-
result = eval(result)
|
314 |
-
except SyntaxError:
|
315 |
-
result = []
|
316 |
-
|
317 |
-
results.extend(result)
|
318 |
-
idx += 50
|
319 |
-
if idx not in df.index:
|
320 |
-
break
|
321 |
-
|
322 |
-
df = pd.DataFrame(results)
|
323 |
-
df = df.merge(df_no_llm.head(1).drop(['Genes', 'SNPs', 'Diseases'], axis=1), 'cross')
|
324 |
-
|
325 |
-
return df, df_no_llm, df_clean
|
|
|
|
|
1 |
from dotenv import load_dotenv
|
2 |
from img2table.document import Image
|
3 |
from langchain.chains.combine_documents.map_reduce import MapReduceDocumentsChain
|
|
|
8 |
from langchain_google_genai import ChatGoogleGenerativeAI
|
9 |
from langchain_openai import ChatOpenAI
|
10 |
from pdf2image import convert_from_path
|
11 |
+
from prompt import *
|
12 |
from table_detector import detection_transform, device, model, ocr, outputs_to_objects
|
13 |
|
14 |
import io
|
|
|
16 |
import os
|
17 |
import pandas as pd
|
18 |
import re
|
|
|
|
|
19 |
import torch
|
20 |
|
21 |
load_dotenv()
|
|
|
28 |
|
29 |
class Process():
|
30 |
|
31 |
+
def __init__(self, llm):
|
32 |
|
33 |
if llm.startswith('gpt'):
|
34 |
self.llm = ChatOpenAI(temperature=0, model_name=llm)
|
|
|
37 |
else:
|
38 |
self.llm = ChatOpenAI(temperature=0, model_name=llm, api_key=os.environ['PERPLEXITY_API_KEY'], base_url="https://api.perplexity.ai")
|
39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
def get_entity(self, data):
|
41 |
|
42 |
chunks, types = data
|
|
|
87 |
|
88 |
def get_table(self, path):
|
89 |
|
|
|
90 |
images = convert_from_path(path)
|
|
|
91 |
tables = []
|
92 |
|
93 |
# Loop pages
|
|
|
112 |
print(detected_tables[idx])
|
113 |
tables.append(cropped_table)
|
114 |
|
|
|
115 |
genes = []
|
116 |
snps = []
|
117 |
diseases = []
|
|
|
134 |
for extracted_table in extracted_tables[1:]:
|
135 |
df_table = pd.concat([df_table, extracted_table.df]).reset_index(drop=True)
|
136 |
|
137 |
+
df_table = df_table.fillna('')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
# Ask LLM with JSON data
|
140 |
+
json_table = df_table.to_json(orient='records')
|
141 |
str_json_table = json.dumps(json.loads(json_table), indent=2)
|
142 |
|
143 |
result = self.llm.invoke(prompt_table.format(str_json_table)).content
|
|
|
159 |
snps.append(snp)
|
160 |
diseases.append(res_disease)
|
161 |
|
|
|
162 |
print(genes, snps, diseases)
|
|
|
163 |
return genes, snps, diseases
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|