add reader
Browse files- app.py +2 -1
- appStore/reader.py +89 -0
- paramconfig.cfg +11 -1
- utils/reader_qa.py +110 -0
app.py
CHANGED
@@ -6,6 +6,7 @@ import appStore.ghg as ghg
|
|
6 |
import appStore.policyaction as policyaction
|
7 |
import appStore.conditional as conditional
|
8 |
import appStore.indicator as indicator
|
|
|
9 |
import appStore.doc_processing as processing
|
10 |
from utils.uploadAndExample import add_upload
|
11 |
import streamlit as st
|
@@ -88,7 +89,7 @@ with st.expander("ℹ️ - About this app", expanded=False):
|
|
88 |
|
89 |
st.write("")
|
90 |
apps = [processing.app, target_extraction.app, netzero.app, ghg.app,
|
91 |
-
policyaction.app, conditional.app, sector.app, adapmit.app,indicator.app]
|
92 |
#conditional.app, sector.app]
|
93 |
#adapmit.app]
|
94 |
|
|
|
6 |
import appStore.policyaction as policyaction
|
7 |
import appStore.conditional as conditional
|
8 |
import appStore.indicator as indicator
|
9 |
+
import appStore.reader as reader
|
10 |
import appStore.doc_processing as processing
|
11 |
from utils.uploadAndExample import add_upload
|
12 |
import streamlit as st
|
|
|
89 |
|
90 |
st.write("")
|
91 |
apps = [processing.app, target_extraction.app, netzero.app, ghg.app,
|
92 |
+
policyaction.app, conditional.app, sector.app, adapmit.app,indicator.app, reader.app]
|
93 |
#conditional.app, sector.app]
|
94 |
#adapmit.app]
|
95 |
|
appStore/reader.py
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# set path
|
2 |
+
import glob, os, sys;
|
3 |
+
sys.path.append('../utils')
|
4 |
+
|
5 |
+
#import needed libraries
|
6 |
+
import seaborn as sns
|
7 |
+
import matplotlib.pyplot as plt
|
8 |
+
import numpy as np
|
9 |
+
import pandas as pd
|
10 |
+
import streamlit as st
|
11 |
+
from utils.reader_qa import load_reader, reader_highlight
|
12 |
+
import logging
|
13 |
+
logger = logging.getLogger(__name__)
|
14 |
+
from utils.config import get_classifier_params
|
15 |
+
from io import BytesIO
|
16 |
+
import xlsxwriter
|
17 |
+
import plotly.express as px
|
18 |
+
|
19 |
+
|
20 |
+
# Declare all the necessary variables
|
21 |
+
classifier_identifier = 'reader'
|
22 |
+
params = get_classifier_params(classifier_identifier)
|
23 |
+
|
24 |
+
|
25 |
+
def app():
|
26 |
+
### Main app code ###
|
27 |
+
with st.container():
|
28 |
+
if 'key1' in st.session_state:
|
29 |
+
df = st.session_state.key1
|
30 |
+
|
31 |
+
# Load the classifier model
|
32 |
+
classifier = load_reader(classifier_name=params['model_name'])
|
33 |
+
st.session_state['{}_qa'.format(classifier_identifier)] = classifier
|
34 |
+
|
35 |
+
if sum(df['Target Label'] == 'TARGET') > 100:
|
36 |
+
warning_msg = ": This might take sometime, please sit back and relax."
|
37 |
+
else:
|
38 |
+
warning_msg = ""
|
39 |
+
|
40 |
+
reader_highlight(haystack_doc=df,
|
41 |
+
threshold= params['threshold'])
|
42 |
+
# st.session_state.key1 = df
|
43 |
+
|
44 |
+
|
45 |
+
|
46 |
+
# @st.cache_data
|
47 |
+
# def to_excel(df):
|
48 |
+
# len_df = len(df)
|
49 |
+
# output = BytesIO()
|
50 |
+
# writer = pd.ExcelWriter(output, engine='xlsxwriter')
|
51 |
+
# df.to_excel(writer, index=False, sheet_name='Sheet1')
|
52 |
+
# workbook = writer.book
|
53 |
+
# worksheet = writer.sheets['Sheet1']
|
54 |
+
# worksheet.data_validation('E2:E{}'.format(len_df),
|
55 |
+
# {'validate': 'list',
|
56 |
+
# 'source': ['No', 'Yes', 'Discard']})
|
57 |
+
# writer.save()
|
58 |
+
# processed_data = output.getvalue()
|
59 |
+
# return processed_data
|
60 |
+
|
61 |
+
# def netzero_display():
|
62 |
+
# if 'key1' in st.session_state:
|
63 |
+
# df = st.session_state.key2
|
64 |
+
# hits = df[df['Netzero Label'] == 'NETZERO']
|
65 |
+
# range_val = min(5,len(hits))
|
66 |
+
# if range_val !=0:
|
67 |
+
# count_df = df['Netzero Label'].value_counts()
|
68 |
+
# count_df = count_df.rename('count')
|
69 |
+
# count_df = count_df.rename_axis('Netzero Label').reset_index()
|
70 |
+
# count_df['Label_def'] = count_df['Netzero Label'].apply(lambda x: _lab_dict[x])
|
71 |
+
|
72 |
+
# fig = px.bar(count_df, y="Label_def", x="count", orientation='h', height =200)
|
73 |
+
# c1, c2 = st.columns([1,1])
|
74 |
+
# with c1:
|
75 |
+
# st.plotly_chart(fig,use_container_width= True)
|
76 |
+
|
77 |
+
# hits = hits.sort_values(by=['Netzero Score'], ascending=False)
|
78 |
+
# st.write("")
|
79 |
+
# st.markdown("###### Top few NetZero Target Classified paragraph/text results ######")
|
80 |
+
# range_val = min(5,len(hits))
|
81 |
+
# for i in range(range_val):
|
82 |
+
# # the page number reflects the page that contains the main paragraph
|
83 |
+
# # according to split limit, the overlapping part can be on a separate page
|
84 |
+
# st.write('**Result {}** `page {}` (Relevancy Score: {:.2f})'.format(i+1,hits.iloc[i]['page'],hits.iloc[i]['Netzero Score']))
|
85 |
+
# st.write("\t Text: \t{}".format(hits.iloc[i]['text']))
|
86 |
+
# else:
|
87 |
+
# st.info("🤔 No Netzero target found")
|
88 |
+
|
89 |
+
|
paramconfig.cfg
CHANGED
@@ -33,7 +33,7 @@ THRESHOLD = 0.50
|
|
33 |
MODEL = ppsingh/mpnet-multilabel-sector-classifier
|
34 |
SPLIT_BY = word
|
35 |
REMOVE_PUNC = 0
|
36 |
-
SPLIT_LENGTH =
|
37 |
SPLIT_OVERLAP = 10
|
38 |
RESPECT_SENTENCE_BOUNDARY = 1
|
39 |
TOP_KEY = 10
|
@@ -86,4 +86,14 @@ REMOVE_PUNC = 0
|
|
86 |
SPLIT_LENGTH = 80
|
87 |
SPLIT_OVERLAP = 10
|
88 |
RESPECT_SENTENCE_BOUNDARY = 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
TOP_KEY = 10
|
|
|
33 |
MODEL = ppsingh/mpnet-multilabel-sector-classifier
|
34 |
SPLIT_BY = word
|
35 |
REMOVE_PUNC = 0
|
36 |
+
SPLIT_LENGTH = 80
|
37 |
SPLIT_OVERLAP = 10
|
38 |
RESPECT_SENTENCE_BOUNDARY = 1
|
39 |
TOP_KEY = 10
|
|
|
86 |
SPLIT_LENGTH = 80
|
87 |
SPLIT_OVERLAP = 10
|
88 |
RESPECT_SENTENCE_BOUNDARY = 1
|
89 |
+
TOP_KEY = 10
|
90 |
+
|
91 |
+
[reader]
|
92 |
+
THRESHOLD = 0.50
|
93 |
+
MODEL = ppsingh/roberta-finetuned-qa-policy_0.1
|
94 |
+
SPLIT_BY = word
|
95 |
+
REMOVE_PUNC = 0
|
96 |
+
SPLIT_LENGTH = 80
|
97 |
+
SPLIT_OVERLAP = 10
|
98 |
+
RESPECT_SENTENCE_BOUNDARY = 1
|
99 |
TOP_KEY = 10
|
utils/reader_qa.py
ADDED
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import List, Tuple
|
2 |
+
from typing_extensions import Literal
|
3 |
+
import logging
|
4 |
+
import pandas as pd
|
5 |
+
from pandas import DataFrame, Series
|
6 |
+
from utils.config import getconfig
|
7 |
+
from utils.preprocessing import processingpipeline
|
8 |
+
import streamlit as st
|
9 |
+
from transformers import pipeline
|
10 |
+
|
11 |
+
|
12 |
+
@st.cache_resource
|
13 |
+
def load_reader(config_file:str = None, classifier_name:str = None):
|
14 |
+
"""
|
15 |
+
loads the document classifier using haystack, where the name/path of model
|
16 |
+
in HF-hub as string is used to fetch the model object.Either configfile or
|
17 |
+
model should be passed.
|
18 |
+
1. https://docs.haystack.deepset.ai/reference/document-classifier-api
|
19 |
+
2. https://docs.haystack.deepset.ai/docs/document_classifier
|
20 |
+
Params
|
21 |
+
--------
|
22 |
+
config_file: config file path from which to read the model name
|
23 |
+
classifier_name: if modelname is passed, it takes a priority if not \
|
24 |
+
found then will look for configfile, else raise error.
|
25 |
+
Return: document classifier model
|
26 |
+
"""
|
27 |
+
if not classifier_name:
|
28 |
+
if not config_file:
|
29 |
+
logging.warning("Pass either model name or config file")
|
30 |
+
return
|
31 |
+
else:
|
32 |
+
config = getconfig(config_file)
|
33 |
+
classifier_name = config.get('reader','MODEL')
|
34 |
+
|
35 |
+
logging.info("Loading Reader")
|
36 |
+
# we are using the pipeline as the model is multilabel and DocumentClassifier
|
37 |
+
# from Haystack doesnt support multilabel
|
38 |
+
# in pipeline we use 'sigmoid' to explicitly tell pipeline to make it multilabel
|
39 |
+
# if not then it will automatically use softmax, which is not a desired thing.
|
40 |
+
# doc_classifier = TransformersDocumentClassifier(
|
41 |
+
# model_name_or_path=classifier_name,
|
42 |
+
# task="text-classification",
|
43 |
+
# top_k = None)
|
44 |
+
|
45 |
+
qa_model = pipeline("question-answering", model=classifier_name )
|
46 |
+
|
47 |
+
return qa_model
|
48 |
+
|
49 |
+
|
50 |
+
@st.cache_data
|
51 |
+
def reader_highlight(haystack_doc:pd.DataFrame,
|
52 |
+
threshold:float = 0.5,
|
53 |
+
classifier_model:pipeline= None
|
54 |
+
)->Tuple[DataFrame,Series]:
|
55 |
+
"""
|
56 |
+
Text-Classification on the list of texts provided. Classifier provides the
|
57 |
+
most appropriate label for each text. these labels are in terms of if text
|
58 |
+
belongs to which particular Sustainable Devleopment Goal (SDG).
|
59 |
+
Params
|
60 |
+
---------
|
61 |
+
haystack_doc: List of haystack Documents. The output of Preprocessing Pipeline
|
62 |
+
contains the list of paragraphs in different format,here the list of
|
63 |
+
Haystack Documents is used.
|
64 |
+
threshold: threshold value for the model to keep the results from classifier
|
65 |
+
classifiermodel: you can pass the classifier model directly,which takes priority
|
66 |
+
however if not then looks for model in streamlit session.
|
67 |
+
In case of streamlit avoid passing the model directly.
|
68 |
+
Returns
|
69 |
+
----------
|
70 |
+
df: Dataframe
|
71 |
+
"""
|
72 |
+
logging.info("Working on Reader")
|
73 |
+
haystack_doc['Extracted Text'] = 'NA'
|
74 |
+
df1 = haystack_doc[haystack_doc['Target Label'] == 'TARGET']
|
75 |
+
df1 = df1.reset_index(drop=True)
|
76 |
+
df = haystack_doc[haystack_doc['Target Label'] == 'NEGATIVE']
|
77 |
+
df = df.reset_index(drop=True)
|
78 |
+
if not classifier_model:
|
79 |
+
reader_model = st.session_state['reader_qa']
|
80 |
+
ques_ = ['What Target/commitments have been made ?'] * len(df1)
|
81 |
+
|
82 |
+
predictions = reader_model(ques_, list(df1.text))
|
83 |
+
|
84 |
+
st.write(predictions)
|
85 |
+
# # getting the sector label and scores
|
86 |
+
# list_ = []
|
87 |
+
# for i in range(len(predictions)):
|
88 |
+
|
89 |
+
# temp = predictions[i]
|
90 |
+
# placeholder = {}
|
91 |
+
# for j in range(len(temp)):
|
92 |
+
# placeholder[temp[j]['label']] = temp[j]['score']
|
93 |
+
# list_.append(placeholder)
|
94 |
+
# labels_ = [{**list_[l]} for l in range(len(predictions))]
|
95 |
+
# truth_df = DataFrame.from_dict(labels_)
|
96 |
+
# truth_df = truth_df.round(2)
|
97 |
+
# # based on threshold value, we convert each sector score into boolean
|
98 |
+
# truth_df = truth_df.astype(float) >= threshold
|
99 |
+
# truth_df = truth_df.astype(str)
|
100 |
+
# # collecting list of Sector Labels
|
101 |
+
# categories = list(truth_df.columns)
|
102 |
+
# # we collect the Sector Labels as set, None represent the value at the index
|
103 |
+
# # in the list of Sector Labels.
|
104 |
+
# truth_df['Sector Label'] = truth_df.apply(lambda x: {i if x[i]=='True' else
|
105 |
+
# None for i in categories}, axis=1)
|
106 |
+
# # we keep all Sector label except None
|
107 |
+
# truth_df['Sector Label'] = truth_df.apply(lambda x: list(x['Sector Label']
|
108 |
+
# -{None}),axis=1)
|
109 |
+
# haystack_doc['Sector Label'] = list(truth_df['Sector Label'])
|
110 |
+
# return haystack_doc
|