98 lines
3.0 KiB
Python
98 lines
3.0 KiB
Python
|
import gensim.downloader
|
||
|
|
import torch.optim as optim
|
||
|
|
import torch.nn as nn
|
||
|
|
import torch
|
||
|
|
import numpy as np
|
||
|
|
from Net import NNet
|
||
|
|
#from timeit import default_timer as timer
|
||
|
|
|
||
|
|
|
||
|
|
def read_data(folder_name):
|
||
|
|
with open(f'{folder_name}/in.tsv', encoding='utf-8') as file:
|
||
|
|
x = [line.lower().split()[:-2] for line in file.readlines()]
|
||
|
|
|
||
|
|
with open(f'{folder_name}/expected.tsv', encoding='utf-8') as file:
|
||
|
|
y = [int(line.split()[0]) for line in file.readlines()]
|
||
|
|
return x, y
|
||
|
|
|
||
|
|
|
||
|
|
def process_data(data, word2vec):
|
||
|
|
processed_data = []
|
||
|
|
for reddit in data:
|
||
|
|
words_sim = [word2vec[word] for word in reddit if word in word2vec]
|
||
|
|
processed_data.append(np.mean(words_sim or [np.zeros(100)], axis=0))
|
||
|
|
return processed_data
|
||
|
|
|
||
|
|
|
||
|
|
def predict(folder_name, model, word_vec):
|
||
|
|
with open(f'{folder_name}/in.tsv', encoding='utf-8') as file:
|
||
|
|
x_data = [line.lower().split()[:-2] for line in file.readlines()]
|
||
|
|
|
||
|
|
x_train = process_data(x_data, word_vec)
|
||
|
|
|
||
|
|
y_predictions = []
|
||
|
|
with torch.no_grad():
|
||
|
|
for i, inputs in enumerate(x_train):
|
||
|
|
inputs = torch.tensor(inputs.astype(np.float32)).to(device)
|
||
|
|
|
||
|
|
y_predicted = model(inputs)
|
||
|
|
y_predictions.append(y_predicted > 0.5)
|
||
|
|
return y_predictions
|
||
|
|
|
||
|
|
|
||
|
|
def save_predictions(folder_name, predicted_labels):
|
||
|
|
predictions = []
|
||
|
|
for pred in predicted_labels:
|
||
|
|
predictions.append(pred.int()[0].item())
|
||
|
|
|
||
|
|
with open(f"{folder_name}/out.tsv", "w", encoding="UTF-8") as file_out:
|
||
|
|
for pred in predictions:
|
||
|
|
file_out.writelines(f"{str(pred)}\n")
|
||
|
|
|
||
|
|
|
||
|
|
device = "cuda" if torch.cuda.is_available() else "cpu"
|
||
|
|
print(device) #gpu is a bit faster here
|
||
|
|
|
||
|
|
word_vectors = gensim.downloader.load("glove-wiki-gigaword-100")
|
||
|
|
|
||
|
|
x_data, y_train = read_data('train')
|
||
|
|
x_train = process_data(x_data, word_vectors)
|
||
|
|
|
||
|
|
model = NNet().to(device)
|
||
|
|
|
||
|
|
criterion = nn.BCELoss()
|
||
|
|
optimizer = optim.SGD(model.parameters(), lr=0.005) #, momentum=0.9)
|
||
|
|
|
||
|
|
for epoch in range(2):
|
||
|
|
running_loss = 0.0
|
||
|
|
correct = 0.
|
||
|
|
total = 0.
|
||
|
|
for i, (inputs, label) in enumerate(zip(x_train, y_train)):
|
||
|
|
inputs = torch.tensor(inputs.astype(np.float32)).to(device)
|
||
|
|
label = torch.tensor(np.array(label).astype(np.float32)).reshape(1).to(device)
|
||
|
|
|
||
|
|
# zero the parameter gradients
|
||
|
|
optimizer.zero_grad()
|
||
|
|
|
||
|
|
# forward + backward + optimize
|
||
|
|
y_predicted = model(inputs)
|
||
|
|
loss = criterion(y_predicted, label)
|
||
|
|
loss.backward()
|
||
|
|
optimizer.step()
|
||
|
|
|
||
|
|
# print statistics
|
||
|
|
running_loss += loss.item()
|
||
|
|
correct += ((y_predicted > 0.5) == label).type(torch.float).sum().item()
|
||
|
|
total += label.size(0)
|
||
|
|
|
||
|
|
if i % 10000 == 9999: # print every 10000 mini-batches
|
||
|
|
print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 10000:.3f}')
|
||
|
|
print(f'Accuracy score: {100 * correct / total} %')
|
||
|
|
running_loss = 0.0
|
||
|
|
|
||
|
|
predicted = predict('dev-0', model, word_vectors)
|
||
|
|
save_predictions('dev-0', predicted)
|
||
|
|
|
||
|
|
predicted = predict('test-A', model, word_vectors)
|
||
|
|
save_predictions('test-A', predicted)
|