99 lines
3.2 KiB
Python
99 lines
3.2 KiB
Python
import pandas as pd
|
|
import numpy as np
|
|
import torch
|
|
from nltk.tokenize import word_tokenize
|
|
import gensim.downloader
|
|
from csv import QUOTE_NONE
|
|
|
|
|
|
print('initialization')
|
|
word2vec = gensim.downloader.load('word2vec-google-news-300')
|
|
|
|
|
|
def get_word2vec(document):
|
|
return np.mean([word2vec[token] for token in document if token in word2vec] or [np.zeros(300)], axis=0)
|
|
|
|
|
|
class MyNeuralNetwork(torch.nn.Module):
|
|
|
|
def __init__(self, input_size, hidden_size, num_classes):
|
|
super(MyNeuralNetwork, self).__init__()
|
|
self.fc1 = torch.nn.Linear(input_size, hidden_size)
|
|
self.fc2 = torch.nn.Linear(hidden_size, num_classes)
|
|
|
|
def forward(self, x):
|
|
x = self.fc1(x)
|
|
x = torch.relu(x)
|
|
x = self.fc2(x)
|
|
x = torch.sigmoid(x)
|
|
return x
|
|
|
|
|
|
# wczytanie danych
|
|
print('loading data')
|
|
train_x = pd.read_table('train/in.tsv.xz', error_bad_lines=False, header=None, quoting=QUOTE_NONE,
|
|
names=['content', 'id'])
|
|
train_y = pd.read_table('train/expected.tsv', error_bad_lines=False, header=None, quoting=QUOTE_NONE,
|
|
names=['label'])['label']
|
|
dev_x = pd.read_table('dev-0/in.tsv.xz', error_bad_lines=False, header=None, quoting=QUOTE_NONE,
|
|
names=['content', 'id'])
|
|
test_x = pd.read_table('test-A/in.tsv.xz', error_bad_lines=False, header=None, quoting=QUOTE_NONE,
|
|
names=['content', 'id'])
|
|
|
|
# preprocessing danych
|
|
print('word tokenize')
|
|
train_x = [word_tokenize(row) for row in train_x['content'].str.lower()]
|
|
dev_x = [word_tokenize(row) for row in dev_x['content'].str.lower()]
|
|
test_x = [word_tokenize(row) for row in test_x['content'].str.lower()]
|
|
|
|
print('word2vec')
|
|
train_x = [get_word2vec(document) for document in train_x]
|
|
dev_x = [get_word2vec(document) for document in dev_x]
|
|
test_x = [get_word2vec(document) for document in test_x]
|
|
|
|
# trenowanie
|
|
print('model training')
|
|
network = MyNeuralNetwork(300, 600, 1)
|
|
criterion = torch.nn.BCELoss()
|
|
optimizer = torch.optim.SGD(network.parameters(), lr=0.02)
|
|
epochs = 15
|
|
batch_size = 15
|
|
|
|
for epoch in range(epochs):
|
|
network.train()
|
|
for i in range(0, train_y.shape[0], batch_size):
|
|
x = train_x[i:i + batch_size]
|
|
x = torch.tensor(x)
|
|
y = train_y[i:i + batch_size]
|
|
y = torch.tensor(y.astype(np.float32).to_numpy()).reshape(-1, 1)
|
|
outputs = network(x.float())
|
|
loss = criterion(outputs, y)
|
|
optimizer.zero_grad()
|
|
loss.backward()
|
|
optimizer.step()
|
|
|
|
# ewaluacja
|
|
print('evaluation')
|
|
dev_y_prediction = []
|
|
test_y_prediction = []
|
|
with torch.no_grad():
|
|
for i in range(0, len(dev_x), batch_size):
|
|
x = dev_x[i:i + batch_size]
|
|
x = torch.tensor(x)
|
|
outputs = network(x.float())
|
|
prediction = outputs > 0.5
|
|
dev_y_prediction += prediction.tolist()
|
|
for i in range(0, len(test_x), batch_size):
|
|
x = test_x[i:i + batch_size]
|
|
x = torch.tensor(x)
|
|
outputs = network(x.float())
|
|
prediction = outputs > 0.5
|
|
test_y_prediction += prediction.tolist()
|
|
|
|
# zapisanie danych
|
|
print('saving data')
|
|
np.asarray(dev_y_prediction, dtype=np.int32).tofile('./dev-0/out.tsv', sep='\n')
|
|
np.asarray(test_y_prediction, dtype=np.int32).tofile('./test-A/out.tsv', sep='\n')
|
|
|
|
print('done')
|