Linker_Fr.py

import pandas as pd
import datetime
import numpy as np
import datetime
import pytz
import warnings
warnings.filterwarnings('ignore')
import argparse

#this function put on the good format the date from the time series dataset
def to_datetime(my_date,set_hour = False):
	my_date = str(my_date)
	set_month= {'Jan': '01', 'Feb' : '02', 'Mar' : '03', 'Apr' : '04','May' : '05', 'Jun' : '06', 'Jul' : '07', 'Aug' : '08', 'Sep' : '09', 'Oct' : '10', 'Nov' : '11', 'Dec' : '12'}
	month = int(my_date[4:6])
	day = int(my_date[6:8])
	year = int(my_date[0:4])
	if set_hour :
		hour = 0
	else :
		hour = int(my_date[8:10])
	unaware = datetime.datetime(year, month, day, hour)
	final_date = pytz.utc.localize(unaware)
	return final_date

def to_datetimeNLP(my_date,set_hour = False):
	my_date = str(my_date)
	set_month= {'Jan': '01', 'Feb' : '02', 'Mar' : '03', 'Apr' : '04','May' : '05', 'Jun' : '06', 'Jul' : '07', 'Aug' : '08', 'Sep' : '09', 'Oct' : '10', 'Nov' : '11', 'Dec' : '12'}
	month = int(my_date[5:7])
	day = int(my_date[8:10])
	year = int(my_date[0:4])
	if set_hour :
		hour = 0
	else :
		hour = int(my_date[11:13])
		hour = hour - (hour%3)
	unaware = datetime.datetime(year, month, day,hour)
	final_date = pytz.utc.localize(unaware)
	return final_date


#the function take a time series dataframe (df) and return an other dataset with all the data put in one parameter (Window) except the date and the label wich are in different columns
#window size is equal to the size of the window
def window_creation(df,size_window,timestamp_colum):
	good_row = list(df.columns)
	good_row.remove(timestamp_colum)
	window_df = df[good_row]
	window_df = window_df.reset_index()
	window_df = window_df.drop(columns=['index'])
	window_df = window_df.tail(size_window)
	window_data = window_df.to_numpy()
	return window_data

#the function that link the NLP data and the time series data for French dataset
#text_data is the french NLP dataset on dataframe format
#crisis_knowledge is a dataframe from the crisis_knowledge CSV
#time_data is the path to the repertory with time series data
#window size is equal to the size of the window
#fillNA is the value to fill the missing value
def linker(text_data : object,crisis_knowledge : object,my_features : str,window_size : int,label_column_1 : str,label_column_2 : str, date_column : str, fillNA : int) -> object :

	#we join crisis knowledge and NLP data on the name of the crisis
	text_data = text_data.join(crisis_knowledge.set_index('Crisis Name'), on='event', validate='m:1')
	text_data['date']  = list(map(to_datetimeNLP,list(text_data['created_at'])))
	#features to keep for Time series
	features =['numer_sta','date','pmer','tend','ff','t','u','n']
	features_clean =['pmer','tend','ff','t','u','n']
	df_return = pd.DataFrame({'Date' : [],'Text' : [], 'Window' : [], 'label' : []})
	list_of_date = []
	list_of_text = []
	list_of_window = []
	list_of_crisis_type = []
	list_of_label_uti = []
	list_of_label_urg = []
	list_of_label_int = []
	list_of_label = []
	#replace missing values and replace label by number
	my_features = my_features[features].fillna(fillNA).replace({'mq': fillNA, 'No_Crisis': 0, 'Crisis': 1, 'Ecological': 1, 'Sudden': 2})
	for clean_f in features_clean :
		my_features[clean_f] = pd.to_numeric(my_features[clean_f])
		my_features[clean_f] = (my_features[clean_f] - my_features[clean_f].min()) / (my_features[clean_f].max() - my_features[clean_f].min())
	#this take from time series data the list of the id for each station
	set_of_station = set(list(my_features['numer_sta']))
	my_features['date'] = list(map(to_datetime,list(my_features['date'])))
	set_of_date = set(list(my_features['date']))
	dict_by_station = {}
	i = 0
	for index,text_line in text_data.iterrows():

		#if the NLP date can be found in the time series date
		if text_line['date'] in set_of_date :

			list_of_station = eval(text_line['Related station'])
			for i in range(len(list_of_station)):
				list_of_station[i] = int(list_of_station[i])
				#station related to the crisis

			current_station = my_features[my_features['numer_sta'] == list_of_station[0]]
			current_station = current_station.sort_values(by=['date'])
			good_date = current_station[current_station['date'] < text_line['date']]
			list_of_date.append(text_line['date'])
			list_of_text.append(text_line['text'])
			list_of_crisis_type.append(text_line['type_crisis'])
			window = window_creation(good_date,window_size,'date')
			list_of_window.append(window)
			list_of_label_uti.append(text_line['utility'])
			list_of_label_int.append(text_line['humanitarian'])
			list_of_label_urg.append(text_line['urgency'])
			if text_line['utility'] == 'Message-NonUtilisable' :
				list_of_label.append('Not_Crisis_period')
			else:
				if text_line['type_crisis'] == 'Flood' or text_line['type_crisis'] == 'Hurricane' or text_line['type_crisis'] == 'Storms':
					list_of_label.append('Ecological_crisis')
				else:
					list_of_label.append('Sudden_Crisis')

	df_return = pd.DataFrame({'Date' : list_of_date,'Text' : list_of_text,'Crisis_Type' : list_of_crisis_type , 'Window' : list_of_window, 'intention' : list_of_label_int, 'urgency' : list_of_label_urg, 'utility' : list_of_label_uti , 'label' : list_of_label})
	return df_return

parser = argparse.ArgumentParser()
parser.add_argument("-w", "--window_size",type = int, default = 16, help="size (in number of days) of our daily time series data")
parser.add_argument("-o", "--output_file",type = str, default = './Multi_modal_dataset/CrisisTS_FR_Personnalized.csv', help="name of your output file")
args = parser.parse_args()

Knowledge = pd.read_csv('./utils/crisis_knowledge_fr.csv',sep="\t")

nlp_csv = pd.read_csv('./Textual_Data/French_Corpus/Kozlwoski.csv',sep ='\t')

time_series = pd.read_csv('./Time_Series/FrenchTS/french_time_series_MERGED.csv',sep = '\t')

nlp_csv = nlp_csv.dropna(subset=['date'])

nlp_csv = nlp_csv.dropna(subset=['humanitarian'])

multi_modal_dataset = linker(nlp_csv ,Knowledge ,time_series , args.window_size,, 'Crisis_Predictability', 'label', 'date', 0)


print('Done !')
multi_modal_test_data.to_csv(args.output_file,sep = '\t')