en-ner-conll-2003/main_updated.ipynb

import numpy as np
import gensim
import torch
import pandas as pd

from torchtext.vocab import Vocab
from collections import Counter

import lzma
import re
import itertools

class NeuralNetworkModel(torch.nn.Module):

    def __init__(self, output_size):
        super(NeuralNetworkModel, self).__init__()
        self.fc1 = torch.nn.Linear(10_000,len(train_tokens_ids))
        self.softmax = torch.nn.Softmax(dim=0)
        

    def forward(self, x):
        x = self.fc1(x)
        x = self.softmax(x)
        return x

class NERModel(torch.nn.Module):

    def __init__(self,):
        super(NERModel, self).__init__()
        self.emb = torch.nn.Embedding(23627,200)
        self.fc1 = torch.nn.Linear(600,9)

    def forward(self, x):
        x = self.emb(x)
        x = x.reshape(600) 
        x = self.fc1(x)
        return x

def get_dataset(path):
    data = lzma.open(path).read().decode('UTF-8').split('\n')
    return [line.split('\t') for line in data][:-1]

train_data = get_dataset('train.tsv.xz')

tokens = []
ner_tags = []

for i in train_data:
    ner_tags.append(i[0].split())
    tokens.append(i[1].split())

ner_tags_set = list(set(itertools.chain(*ner_tags)))

ner_tags_dictionary = {}

for i in range(len(ner_tags_set)):
    ner_tags_dictionary[ner_tags_set[i]] = i

for i in range(len(ner_tags)):
    for j in range(len(ner_tags[i])):
        ner_tags[i][j] = ner_tags_dictionary[ner_tags[i][j]]

def data_preprocessing(data):
    return [ torch.tensor([vocab['<bos>']] +[vocab[token]  for token in  document ] + [vocab['<eos>']], dtype = torch.long) for document in data ]

def labels_preprocessing(data):
    return [ torch.tensor([0] + document + [0], dtype = torch.long) for document in data ]

def build_vocab(dataset):
    counter = Counter()
    for document in dataset:
        counter.update(document)
    return Vocab(counter)


vocab = build_vocab(tokens)
train_tokens_ids = data_preprocessing(tokens)
train_labels = labels_preprocessing(ner_tags)

nn_model = NeuralNetworkModel(len(train_tokens_ids))
train_tokens_ids[0][1:4]

ner_model = NERModel()
ner_model(train_tokens_ids[0][1:4])

criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(ner_model.parameters())

for epoch in range(2):
    loss_score = 0
    acc_score = 0
    prec_score = 0
    selected_items = 0
    recall_score = 0
    relevant_items = 0
    items_total = 0
    nn_model.train()
    for i in range(100):
        for j in range(1, len(train_labels[i]) - 1):
    
            X = train_tokens_ids[i][j-1: j+2]
            Y = train_labels[i][j: j+1]

            Y_predictions = ner_model(X)
            
            
            acc_score += int(torch.argmax(Y_predictions) == Y)
            
            if torch.argmax(Y_predictions) != 0:
                selected_items +=1
            if  torch.argmax(Y_predictions) != 0 and torch.argmax(Y_predictions) == Y.item():
                prec_score += 1
            
            if  Y.item() != 0:
                relevant_items +=1
            if  Y.item() != 0 and torch.argmax(Y_predictions) == Y.item():
                recall_score += 1
            
            items_total += 1

            
            optimizer.zero_grad()
            loss = criterion(Y_predictions.unsqueeze(0), Y)
            loss.backward()
            optimizer.step()


            loss_score += loss.item() 
    
    precision = prec_score / selected_items
    recall = recall_score / relevant_items
    f1_score = (2*precision * recall) / (precision + recall)
    display('epoch: ', epoch)
    display('loss: ', loss_score / items_total)
    display('acc: ', acc_score / items_total)
    display('prec: ', precision)
    display('recall: : ', recall)
    display('f1: ', f1_score)

'epoch: '

0

'loss: '

0.5260595670613091

'acc: '

0.8669271431854746

'prec: '

0.868387037208014

'recall: : '

0.8694707649641784

'f1: '

0.868928563179897

'epoch: '

1

'loss: '

0.46469578519580995

'acc: '

0.8826936336474374

'prec: '

0.8842406612180819

'recall: : '

0.885139819736538

'f1: '

0.884690012011457

def process_output(lines):
    result = []
    for line in lines:
        last_label = None
        new_line = []
        for label in line:
            if(label != "O" and label[0:2] == "I-"):
                if last_label == None or last_label == "O":
                    label = label.replace('I-', 'B-')
                else:
                    label = "I-" + last_label[2:]
            last_label = label
            new_line.append(label)
            x = (" ".join(new_line))
        result.append(" ".join(new_line))
    return result

with open('dev-0/in.tsv', "r", encoding="utf-8") as f:
    dev_0_data = [line.rstrip() for line in f]
    
dev_0_data = [i.split() for i in dev_0_data]

with open('dev-0/expected.tsv', "r", encoding="utf-8") as f:
    dev_0_tags = [line.rstrip() for line in f]
    
dev_0_tags = [i.split() for i in dev_0_tags]

for i in range(len(dev_0_tags)):
    for j in range(len(dev_0_tags[i])):
        dev_0_tags[i][j] = ner_tags_dictionary[dev_0_tags[i][j]]
        
test_tokens_ids = data_preprocessing(dev_0_data)
test_labels = labels_preprocessing(dev_0_tags)

result = []

loss_score = 0
acc_score = 0
prec_score = 0
selected_items = 0
recall_score = 0
relevant_items = 0
items_total = 0
nn_model.eval()

for i in range(len(test_tokens_ids)):
    result.append([])
    for j in range(1, len(test_labels[i]) - 1):

        X = test_tokens_ids[i][j-1: j+2]
        Y = test_labels[i][j: j+1]

        Y_predictions = ner_model(X)


        acc_score += int(torch.argmax(Y_predictions) == Y)

        if torch.argmax(Y_predictions) != 0:
            selected_items +=1
        if  torch.argmax(Y_predictions) != 0 and torch.argmax(Y_predictions) == Y.item():
            prec_score += 1

        if  Y.item() != 0:
            relevant_items +=1
        if  Y.item() != 0 and torch.argmax(Y_predictions) == Y.item():
            recall_score += 1

        items_total += 1
        loss = criterion(Y_predictions.unsqueeze(0), Y)
        loss_score += loss.item() 
        
        result[i].append(int(torch.argmax(Y_predictions)))

precision = prec_score / selected_items
recall = recall_score / relevant_items
f1_score = (2*precision * recall) / (precision + recall)
display('loss: ', loss_score / items_total)
display('acc: ', acc_score / items_total)
display('prec: ', precision)
display('recall: : ', recall)
display('f1: ', f1_score)

'loss: '

0.9558752998502382

'acc: '

0.8221458628103122

'prec: '

0.8261110597800979

'recall: : '

0.8253637102134259

'f1: '

0.8257372158959724

def save_to_file(path, results):
  with open(path, "w") as f:
    for line in results:
      print(line)
      f.write(line + "\n")

tags = []
tmp = []
for i in ner_tags_dictionary:
    tmp.append(i)

for i in range(len(result)):
    tags.append([])
    for j in range(len(result[i])):
        tags[i].append(tmp[result[i][j]])

processed_tags = process_output(tags)

with open("dev-0/out.tsv", "w") as f:
    for line in processed_tags:
      f.write(line + "\n")

with open('dev-0/expected.tsv', "r", encoding="utf-8") as f:
    dev_0_tags = [line.rstrip() for line in f]
    
dev_0_tags = [i.split() for i in dev_0_tags]

with open('test-A/in.tsv', "r", encoding="utf-8") as file:
    test_data = [line.rstrip() for line in file]
    
test_data = [i.split() for i in test_data]
test_tokens_ids = data_preprocessing(test_data)
result = []

loss_score = 0
acc_score = 0
prec_score = 0
selected_items = 0
recall_score = 0
relevant_items = 0
items_total = 0
nn_model.eval()

test_tokens_length = len(test_tokens_ids)

for i in range(test_tokens_length):
    result.append([])
    for j in range(1, len(test_tokens_ids[i]) - 1):
        X = test_tokens_ids[i][j-1: j + 2]
        Y_predictions = ner_model(X)
        result[i].append(int(torch.argmax(Y_predictions)))

tags = []
tmp = []

for i in ner_tags_dictionary:
    tmp.append(i)

result_length = len(result)

for i in range(result_length):
    tags.append([])
    for j in range(len(result[i])):
        tags[i].append(tmp[result[i][j]])

processed_tags = process_output(tags)

with open("test-A/out.tsv", "w") as f:
    for line in processed_tags:
      f.write(line + "\n")