# import all packages import requests import streamlit as st from sklearn.model_selection import StratifiedKFold from sklearn.model_selection import train_test_split from sklearn.model_selection import KFold # tokenizer from transformers import AutoTokenizer, DistilBertTokenizerFast # sequence tagging model + training-related from transformers import DistilBertForTokenClassification, Trainer, TrainingArguments import torch import sys import os from sklearn.metrics import classification_report from pandas import read_csv from sklearn.linear_model import LogisticRegression import sklearn.model_selection from sklearn.feature_extraction.text import TfidfTransformer from sklearn.feature_extraction.text import CountVectorizer from sklearn.pipeline import Pipeline, FeatureUnion import math # from sklearn.metrics import accuracy_score # from sklearn.metrics import precision_recall_fscore_support import json import re import numpy as np import pandas as pd import nltk nltk.download("punkt") import string from sklearn.model_selection import train_test_split from transformers import AutoTokenizer, Trainer, TrainingArguments, AutoModelForSequenceClassification, AutoConfig from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler import itertools from transformers import TextClassificationPipeline, TFAutoModelForSequenceClassification, AutoTokenizer from transformers import pipeline import pickle import csv import pdfplumber import pathlib import shutil import webbrowser from streamlit.components.v1 import html import streamlit.components.v1 as components from PyPDF2 import PdfReader from huggingface_hub import HfApi import io from datasets import load_dataset import time import huggingface_hub from huggingface_hub import Repository from datetime import datetime import pathlib as Path from requests import get import urllib.request # import gradio as gr # from gradio import inputs, outputs from datasets import load_dataset from huggingface_hub import HfApi, list_models import os from huggingface_hub import HfFileSystem from tensorflow.keras.models import Sequential, model_from_json #import tensorflow_datasets as tfds import tensorflow as tf from tensorflow.keras.preprocessing.sequence import pad_sequences import spacy from tensorflow.keras.preprocessing.text import Tokenizer #from spacy import en_core_web_lg #import en_core_web_lg #nlp = en_core_web_lg.load() nlp = spacy.load('en_core_web_sm') #tfds.disable_progress_bar() MAX_SEQUENCE_LENGTH = 500 # dataset = load_dataset('Seetha/Visualization', streaming=True) # df = pd.DataFrame.from_dict(dataset['train']) # DATASET_REPO_URL = "https://huggingface.co/datasets/Seetha/Visualization" # DATA_FILENAME = "level2.json" #DATA_FILE = os.path.join("data", DATA_FILENAME) DATASET_REPO_URL = "https://huggingface.co/datasets/Seetha/visual_files" DATA_FILENAME = "detailedResults.json" DATA_FILENAME1 = "level2.json" HF_TOKEN = os.environ.get("HF_TOKEN") #st.write("is none?", HF_TOKEN is None) def main(): st.title("Text to Causal Knowledge Graph") st.sidebar.title("Please upload your text documents in one file here:") k=2 seed = 1 k1= 5 text_list = [] causal_sents = [] uploaded_file = None try: uploaded_file = st.sidebar.file_uploader("Choose a file", type = "pdf") except: uploaded_file = PdfReader('sample_anno.pdf') st.error("Please upload your own PDF to be analyzed") if uploaded_file is not None: reader = PdfReader(uploaded_file) for page in reader.pages: text = page.extract_text() text_list.append(text) else: st.error("Please upload your own PDF to be analyzed") st.stop() text_list_final = [x.replace('\n', '') for x in text_list] text_list_final = re.sub('"', '', str(text_list_final)) sentences = nltk.sent_tokenize(text_list_final) result =[] for i in sentences: result1 = i.lower() result2 = re.sub(r'[^\w\s]','',result1) result.append(result2) #st.write("--- %s seconds ---" % (time.time() - start_time)) tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased") #bert-base-uncased model_path = "checkpoint-2850" model = AutoModelForSequenceClassification.from_pretrained(model_path,id2label={0:'non-causal',1:'causal'}) #st.write('sequence classification loaded') pipe1 = pipeline("text-classification", model=model,tokenizer=tokenizer) for sent in result: pred = pipe1(sent) for lab in pred: if lab['label'] == 'causal': #causal causal_sents.append(sent) # st.write('causal sentence classification finished') # st.write("--- %s seconds ---" % (time.time() - start_time)) model_name = "distilbert-base-cased" tokenizer = DistilBertTokenizerFast.from_pretrained(model_name,low_cpu_mem_usage=True) model_path1 = "DistilBertforTokenclassification" model = DistilBertForTokenClassification.from_pretrained(model_path1,low_cpu_mem_usage=True) #len(unique_tags),, num_labels= 7, , id2label={0:'CT',1:'E',2:'C',3:'O'} pipe = pipeline('ner', model=model, tokenizer=tokenizer,aggregation_strategy='simple') #grouped_entities=True st.write('DistilBERT loaded') sentence_pred = [] class_list = [] entity_list = [] for k in causal_sents: pred= pipe(k) #st.write(pred) #st.write('preds') for i in pred: sentence_pred.append(k) class_list.append(i['word']) entity_list.append(i['entity_group']) # st.write('causality extraction finished') # st.write("--- %s seconds ---" % (time.time() - start_time)) filename = 'Checkpoint-classification.sav' loaded_model = pickle.load(open(filename, 'rb')) loaded_vectorizer = pickle.load(open('vectorizefile_classification.pickle', 'rb')) pipeline_test_output = loaded_vectorizer.transform(class_list) predicted = loaded_model.predict(pipeline_test_output) # tokenizer = Tokenizer(num_words=100000) # tokenizer.fit_on_texts(class_list) # word_index = tokenizer.word_index # text_embedding = np.zeros((len(word_index) + 1, 300)) # for word, i in word_index.items(): # text_embedding[i] = nlp(word).vector # json_file = open('model.json', 'r') # loaded_model_json = json_file.read() # json_file.close() # loaded_model = model_from_json(loaded_model_json) # # load weights into new model # loaded_model.load_weights("model.h5") # loss = tf.keras.losses.CategoricalCrossentropy() #from_logits=True # loaded_model.compile(loss=loss,optimizer=tf.keras.optimizers.Adam(1e-4)) # predictions = loaded_model.predict(pad_sequences(tokenizer.texts_to_sequences(class_list),maxlen=MAX_SEQUENCE_LENGTH)) # predicted = np.argmax(predictions,axis=1) # st.write(predictions) # st.write(predicted) # st.write('stakeholder taxonomy finished') # st.write("--- %s seconds ---" % (time.time() - start_time)) pred1 = predicted level0 = [] count =0 for i in predicted: if i == 3: level0.append('Non-Performance') count +=1 else: level0.append('Performance') count +=1 list_pred = {0: 'Customers',1:'Employees',2:'Investors',3:'Non-performance',4:'Society',5:'Unclassified'} pred_val = [list_pred[i] for i in pred1] #print('count',count) for ind,(sent,preds) in enumerate(zip(class_list,pred_val)): if 'customers' in sent or 'client' in sent or 'consumer' in sent or 'user' in sent: pred_val[ind] = 'Customers' elif 'investor' in sent or 'finance' in sent or 'shareholder' in sent or 'stockholder' in sent or 'owners' in sent: pred_val[ind] = 'Investors' elif 'employee' in sent or 'worker' in sent or 'staff' in sent: pred_val[ind] = 'Employees' elif 'society' in sent or 'societal' in sent or 'social responsib*' in sent or 'social performance' in sent or 'community' in sent: pred_val[ind] = 'Society' sent_id, unique = pd.factorize(sentence_pred) final_list = pd.DataFrame( {'Id': sent_id, 'Fullsentence': sentence_pred, 'Component': class_list, 'causeOrEffect': entity_list, 'Labellevel1': level0, 'Labellevel2': pred_val }) s = final_list['Component'].shift(-1) m = s.str.startswith('##', na=False) final_list.loc[m, 'Component'] += (' ' + s[m]) final_list1 = final_list[~final_list['Component'].astype(str).str.startswith('##')] li = [] uni = final_list1['Id'].unique() for i in uni: df_new = final_list1[final_list1['Id'] == i] uni1 = df_new['Id'].unique() # if 'E' not in df_new.values: # li.append(uni1) # out = np.concatenate(li).ravel() # li_pan = pd.DataFrame(out,columns=['Id']) # df3 = pd.merge(final_list1, li_pan[['Id']], on='Id', how='left', indicator=True) \ # .query("_merge == 'left_only'") \ # .drop("_merge",axis=1) df3 = final_list1 #df = df3.groupby(['Id','Fullsentence','causeOrEffect', 'Labellevel1', 'Labellevel2'])['Component'].apply(', '.join).reset_index() #st.write(df) #df = df3 df3["causeOrEffect"].replace({"C": "cause", "E": "effect"}, inplace=True) df_final = df3[df3['causeOrEffect'] != 'CT'] df3['New string'] = df_final['Component'].replace(r'[##]+', ' ', regex=True) df_final = df_final.drop("Component",axis=1) df_final.insert(2, "Component", df3['New string'], True) df_final1 = df_final[df_final['Component'].str.split().str.len().gt(1)] #st.write(df_final[df_final['Component'].str.len() != 1]) #df_final1.to_csv('predictions.csv') # buffer = io.BytesIO() # with pd.ExcelWriter(buffer, engine="xlsxwriter") as writer: # df_final.to_excel(writer, sheet_name="Sheet1", index=False) # writer.close() count_NP_NP = 0 count_NP_investor = 0 count_NP_customer = 0 count_NP_employees = 0 count_NP_society = 0 count_inv_np = 0 count_inv_investor = 0 count_inv_customer = 0 count_inv_employee = 0 count_inv_society = 0 count_cus_np = 0 count_cus_investor = 0 count_cus_customer = 0 count_cus_employee = 0 count_cus_society = 0 count_emp_np = 0 count_emp_investor = 0 count_emp_customer = 0 count_emp_employee = 0 count_emp_society = 0 count_soc_np = 0 count_soc_investor = 0 count_soc_customer = 0 count_soc_employee = 0 count_soc_society = 0 for i in range(0,df_final['Id'].max()): j = df_final.loc[df_final['Id'] == i] cause_tab = j.loc[j['causeOrEffect'] == 'cause'] effect_tab = j.loc[j['causeOrEffect'] == 'effect'] cause_coun_NP = (cause_tab.Labellevel2 == 'Non-performance').sum() effect_coun_NP = (effect_tab.Labellevel2 == 'Non-performance').sum() if (cause_coun_NP > 0) and (effect_coun_NP > 0): count_NP = cause_coun_NP if cause_coun_NP >= effect_coun_NP else effect_coun_NP else: count_NP = 0 effect_NP_inv = (effect_tab.Labellevel2 == 'Investors').sum() if (cause_coun_NP > 0) and (effect_NP_inv > 0): count_NP_inv = cause_coun_NP if cause_coun_NP >= effect_NP_inv else effect_NP_inv else: count_NP_inv = 0 effect_NP_cus = (effect_tab.Labellevel2 == 'Customers').sum() if (cause_coun_NP > 0) and (effect_NP_cus > 0): count_NP_cus = cause_coun_NP if cause_coun_NP >= effect_NP_cus else effect_NP_cus else: count_NP_cus = 0 effect_NP_emp = (effect_tab.Labellevel2 == 'Employees').sum() if (cause_coun_NP > 0) and (effect_NP_emp > 0): count_NP_emp = cause_coun_NP if cause_coun_NP >= effect_NP_emp else effect_NP_emp else: count_NP_emp = 0 effect_NP_soc = (effect_tab.Labellevel2 == 'Society').sum() if (cause_coun_NP > 0) and (effect_NP_soc > 0): count_NP_soc = cause_coun_NP if cause_coun_NP >= effect_NP_soc else effect_NP_soc else: count_NP_soc = 0 cause_coun_inv = (cause_tab.Labellevel2 == 'Investors').sum() effect_coun_inv = (effect_tab.Labellevel2 == 'Non-performance').sum() if (cause_coun_inv > 0) and (effect_coun_inv > 0): count_NP_inv = cause_coun_inv if cause_coun_inv >= effect_coun_inv else effect_coun_inv else: count_NP_inv = 0 effect_inv_inv = (effect_tab.Labellevel2 == 'Investors').sum() if (cause_coun_inv > 0) and (effect_inv_inv > 0): count_inv_inv = cause_coun_inv if cause_coun_inv >= effect_inv_inv else effect_inv_inv else: count_inv_inv = 0 effect_inv_cus = (effect_tab.Labellevel2 == 'Customers').sum() if (cause_coun_inv > 0) and (effect_inv_cus > 0): count_inv_cus = cause_coun_inv if cause_coun_inv >= effect_inv_cus else effect_inv_cus else: count_inv_cus = 0 effect_inv_emp = (effect_tab.Labellevel2 == 'Employees').sum() if (cause_coun_inv > 0) and (effect_inv_emp > 0): count_inv_emp = cause_coun_inv if cause_coun_inv >= effect_inv_emp else effect_inv_emp else: count_inv_emp = 0 effect_inv_soc = (effect_tab.Labellevel2 == 'Society').sum() if (cause_coun_inv > 0) and (effect_inv_soc > 0): count_inv_soc = cause_coun_inv if cause_coun_inv >= effect_inv_soc else effect_inv_soc else: count_inv_soc = 0 cause_coun_cus = (cause_tab.Labellevel2 == 'Customers').sum() effect_coun_cus = (effect_tab.Labellevel2 == 'Non-performance').sum() if (cause_coun_cus > 0) and (effect_coun_cus > 0): count_NP_cus = cause_coun_cus if cause_coun_cus >= effect_coun_cus else effect_coun_cus else: count_NP_cus = 0 effect_cus_inv = (effect_tab.Labellevel2 == 'Investors').sum() if (cause_coun_cus > 0) and (effect_cus_inv > 0): count_cus_inv = cause_coun_cus if cause_coun_cus >= effect_cus_inv else effect_cus_inv else: count_cus_inv = 0 effect_cus_cus = (effect_tab.Labellevel2 == 'Customers').sum() if (cause_coun_cus > 0) and (effect_cus_cus > 0): count_cus_cus = cause_coun_cus if cause_coun_cus >= effect_cus_cus else effect_cus_cus else: count_cus_cus = 0 effect_cus_emp = (effect_tab.Labellevel2 == 'Employees').sum() if (cause_coun_cus > 0) and (effect_cus_emp > 0): count_cus_emp = cause_coun_cus if cause_coun_cus >= effect_cus_emp else effect_cus_emp else: count_cus_emp = 0 effect_cus_soc = (effect_tab.Labellevel2 == 'Society').sum() if (cause_coun_cus > 0) and (effect_cus_soc > 0): count_cus_soc = cause_coun_cus if cause_coun_cus >= effect_cus_soc else effect_cus_soc else: count_cus_soc = 0 cause_coun_emp = (cause_tab.Labellevel2 == 'Employees').sum() effect_coun_emp = (effect_tab.Labellevel2 == 'Non-performance').sum() if (cause_coun_emp > 0) and (effect_coun_emp > 0): count_NP_emp = cause_coun_emp if cause_coun_emp >= effect_coun_emp else effect_coun_emp else: count_NP_emp = 0 effect_emp_inv = (effect_tab.Labellevel2 == 'Investors').sum() if (cause_coun_emp > 0) and (effect_emp_inv > 0): count_emp_inv = cause_coun_emp if cause_coun_emp >= effect_emp_inv else effect_emp_inv else: count_emp_inv = 0 effect_emp_cus = (effect_tab.Labellevel2 == 'Customers').sum() if (cause_coun_emp > 0) and (effect_emp_cus > 0): count_emp_cus = cause_coun_emp if cause_coun_emp >= effect_emp_cus else effect_emp_cus else: count_emp_cus = 0 effect_emp_emp = (effect_tab.Labellevel2 == 'Employees').sum() if (cause_coun_emp > 0) and (effect_emp_emp > 0): count_emp_emp = cause_coun_emp if cause_coun_emp >= effect_emp_emp else effect_emp_emp else: count_emp_emp = 0 effect_emp_soc = (effect_tab.Labellevel2 == 'Society').sum() if (cause_coun_emp > 0) and (effect_emp_soc > 0): count_emp_soc = cause_coun_emp if cause_coun_emp >= effect_emp_soc else effect_emp_soc else: count_emp_soc = 0 cause_coun_soc = (cause_tab.Labellevel2 == 'Society').sum() effect_coun_soc = (effect_tab.Labellevel2 == 'Non-performance').sum() if (cause_coun_soc > 0) and (effect_coun_soc > 0): count_NP_soc = cause_coun_soc if cause_coun_soc >= effect_coun_soc else effect_coun_soc else: count_NP_soc = 0 effect_soc_inv = (effect_tab.Labellevel2 == 'Investors').sum() if (cause_coun_soc > 0) and (effect_soc_inv > 0): count_soc_inv = cause_coun_soc if cause_coun_soc >= effect_soc_inv else effect_soc_inv else: count_soc_inv = 0 effect_soc_cus = (effect_tab.Labellevel2 == 'Customers').sum() if (cause_coun_soc > 0) and (effect_soc_cus > 0): count_soc_cus = cause_coun_soc if cause_coun_soc >= effect_soc_cus else effect_soc_cus else: count_soc_cus = 0 effect_soc_emp = (effect_tab.Labellevel2 == 'Employees').sum() if (cause_coun_soc > 0) and (effect_soc_emp > 0): count_soc_emp = cause_coun_soc if cause_coun_soc >= effect_soc_emp else effect_soc_emp else: count_soc_emp = 0 effect_soc_soc = (effect_tab.Labellevel2 == 'Society').sum() if (cause_coun_soc > 0) and (effect_soc_soc > 0): count_soc_soc = cause_coun_soc if cause_coun_soc >= effect_soc_soc else effect_soc_soc else: count_soc_soc = 0 count_NP_NP = count_NP_NP + count_NP count_NP_investor = count_NP_investor + count_NP_inv count_NP_customer = count_NP_customer + count_NP_cus count_NP_employees = count_NP_employees + count_NP_emp count_NP_society = count_NP_society + count_NP_soc count_inv_np = count_inv_np + count_NP_inv count_inv_investor = count_inv_investor + count_inv_inv count_inv_customer = count_inv_customer + count_inv_cus count_inv_employee = count_inv_employee + count_inv_emp count_inv_society = count_inv_society + count_inv_soc count_cus_np = count_cus_np + count_NP_cus count_cus_investor = count_cus_investor + count_cus_inv count_cus_customer = count_cus_customer + count_cus_cus count_cus_employee = count_cus_employee + count_cus_emp count_cus_society = count_cus_society + count_cus_soc count_emp_np = count_emp_np + count_NP_emp count_emp_investor = count_emp_investor + count_emp_inv count_emp_customer = count_emp_customer + count_emp_cus count_emp_employee = count_emp_employee + count_emp_emp count_emp_society = count_emp_society + count_emp_soc count_soc_np = count_soc_np + count_NP_soc count_soc_investor = count_soc_investor + count_soc_inv count_soc_customer = count_soc_customer + count_soc_cus count_soc_employee = count_soc_employee + count_soc_emp count_soc_society = count_soc_society + count_soc_soc df_tab = pd.DataFrame(columns = ['Non-performance', 'Investors', 'Customers', 'Employees', 'Society'],index=['Non-performance', 'Investors', 'Customers', 'Employees', 'Society'], dtype=object) df_tab.loc['Non-performance'] = [count_NP_NP, count_NP_investor, count_NP_customer, count_NP_employees, count_NP_society] df_tab.loc['Investors'] = [count_inv_np, count_inv_investor, count_inv_customer, count_inv_employee, count_inv_society] df_tab.loc['Customers'] = [count_cus_np, count_cus_investor, count_cus_customer, count_cus_employee, count_cus_society] df_tab.loc['Employees'] = [count_emp_np, count_emp_investor, count_emp_customer, count_emp_employee, count_emp_society] df_tab.loc['Society'] = [count_soc_np, count_soc_investor, count_soc_customer, count_soc_employee, count_soc_society] # df_tab = pd.DataFrame({ # 'Non-performance': [count_NP_NP, count_NP_investor, count_NP_customer, count_NP_employees, count_NP_society], # 'Investors': [count_inv_np, count_inv_investor, count_inv_customer, count_inv_employee, count_inv_society], # 'Customers': [count_cus_np, count_cus_investor, count_cus_customer, count_cus_employee, count_cus_society], # 'Employees': [count_emp_np, count_emp_investor, count_emp_customer, count_emp_employee, count_emp_society], # 'Society': [count_soc_np, count_soc_investor, count_soc_customer, count_soc_employee, count_soc_society]}, # index=['Non-performance', 'Investors', 'Customers', 'Employees', 'Society']) #df_tab.to_csv('final_data.csv') buffer = io.BytesIO() with pd.ExcelWriter(buffer, engine="xlsxwriter") as writer: df_tab.to_excel(writer,sheet_name="count_result",index=False) df_final1.to_excel(writer,sheet_name="Detailed_results",index=False) writer.close() #df = pd.read_csv('final_data.csv', index_col=0) #474-515 # Convert to JSON format json_data = [] for row in df_tab.index: for col in df_tab.columns: json_data.append({ 'source': row, 'target': col, 'value': int(df_tab.loc[row, col]) }) HfApi().delete_file(path_in_repo = DATA_FILENAME1 ,repo_id = 'Seetha/visual_files',token= HF_TOKEN,repo_type='dataset') #st.write('file-deleted') fs = HfFileSystem(token=HF_TOKEN) with fs.open('datasets/Seetha/visual_files/level2.json', 'w') as f: json.dump(json_data, f) df_final1.to_csv('predictions.csv') csv_file = "predictions.csv" json_file = "detailedResults.json" # Open the CSV file and read the data with open(csv_file, "r") as f: csv_data = csv.DictReader(f) # # Convert the CSV data to a list of dictionaries data_list = [] for row in csv_data: data_list.append(dict(row)) # # Convert the list of dictionaries to JSON json_data = json.dumps(data_list) HfApi().delete_file(path_in_repo = DATA_FILENAME ,repo_id = 'Seetha/visual_files',token= HF_TOKEN,repo_type='dataset') #st.write('file2-deleted') with fs.open('datasets/Seetha/visual_files/detailedResults.json','w') as fi: #data = json.load(fi) fi.write(json_data) def convert_df(df): #IMPORTANT: Cache the conversion to prevent computation on every rerun return df.to_csv().encode('utf-8') csv1 = convert_df(df_final1.astype(str)) csv2 = convert_df(df_tab.astype(str)) with st.container(): st.download_button(label="Download the result table",data=buffer,file_name="t2cg_outputs.xlsx",mime="application/vnd.ms-excel") st.markdown('Click this link in a separate tab to view knowledge graph', unsafe_allow_html=True) # st.download_button(label="Download the detailed result table_csv",data=csv1,file_name='results.csv',mime='text/csv') # st.download_button(label="Download the result table_csv",data=csv2,file_name='final_data.csv',mime='text/csv') #with st.container(): # Execute your app #st.title("Visualization example") # components.html(source_code) #html(my_html) #webbrowser.open('https://huggingface.co/spaces/Seetha/visual-knowledgegraph') # # embed streamlit docs in a streamlit app # #components.iframe("https://webpages.charlotte.edu/ltotapal/") if __name__ == '__main__': start_time = time.time() main()