import torch.utils.data
import dataset
from echotorch.transforms import text
# Experience parameter
batch_size = 64
n_epoch = 1
window_size = 700
training_set_size = 10
test_set_size = 2
training_samples = training_set_size + test_set_size
stride = 100
# Style change detection dataset, training set
pan18loader_train = torch.utils.data.DataLoader(
dataset.SCDSimpleDataset(root='./data/', download=True, transform=text.Character(), train=True),
batch_size=1
)
# Style change detection dataset, validation set
pan18loader_valid = torch.utils.data.DataLoader(
dataset.SCDSimpleDataset(root='./data/', download=True, transform=text.Character(), train=False),
batch_size=1
)
# Get training data
for i, data in enumerate(pan18loader_train):
# Inputs and c
inputs, label = data
# TRAINING
import torch.utils.data
import dataset
from echotorch.transforms import text
import random
from torch.autograd import Variable
# Experience parameter
window_size = 500
batch_size = 64
sample_batch = 4
epoch_batches = 10
max_epoch = 1
# Author identification training dataset
pan18loader_training = torch.utils.data.DataLoader(
dataset.AuthorIdentificationDataset(root="./data/", download=True, transform=text.Character(), problem=1),
batch_size=1, shuffle=True)
# Author identification test dataset
pan18loader_test = torch.utils.data.DataLoader(
dataset.AuthorIdentificationDataset(root="./data/", download=True, transform=text.Character(), problem=1,
train=False),
batch_size=1, shuffle=True)
# Authors
author_to_idx = dict()
for idx, author in enumerate(pan18loader_training.dataset.authors):
author_to_idx[author] = idx
# end for
# Number of authors
from torch import optim
import torch.nn as nn
import echotorch.nn as etnn
import echotorch.utils
import os
# Settings
n_epoch = 1
embedding_dim = 10
n_authors = 15
use_cuda = True
voc_size = 58
# Word embedding
transform = text.Character()
# Reuters C50 dataset
reutersloader = torch.utils.data.DataLoader(datasets.ReutersC50Dataset(download=True, n_authors=15,
transform=transform),
batch_size=1, shuffle=False)
# Model
model = CNNCharacterEmbedding(voc_size=voc_size, embedding_dim=embedding_dim)
# Optimizer
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)
# Loss function
# loss_function = nn.NLLLoss()
loss_function = nn.CrossEntropyLoss()
import torch.utils.data
import dataset
from echotorch.transforms import text
# Experience parameter
batch_size = 64
n_epoch = 1
window_size = 700
training_set_size = 10
test_set_size = 2
training_samples = training_set_size + test_set_size
stride = 100
# Style change detection dataset, training set
pan18loader_train = torch.utils.data.DataLoader(dataset.SCDSimpleDataset(
root='./data/', download=True, transform=text.Character(), train=True),
batch_size=1)
# Style change detection dataset, validation set
pan18loader_valid = torch.utils.data.DataLoader(dataset.SCDSimpleDataset(
root='./data/', download=True, transform=text.Character(), train=False),
batch_size=1)
# For each epoch
for epoch in range(n_epoch):
# Training loss
training_loss = 0.0
# Get training data
for i, data in enumerate(pan18loader_train):
# Inputs and c
input_sparsity = space['input_sparsity']
spectral_radius = space['spectral_radius']
transformer = space['transformer'][0][0]
aggregation = space['aggregation'][0][0]
state_gram = space['state_gram']
dataset_start = int(space['dataset_start'])
# Choose the right transformer
if "wv" in transformer:
reutersloader.dataset.transform = text.GloveVector()
elif "pos" in transformer:
reutersloader.dataset.transform = text.PartOfSpeech()
elif "tag" in transformer:
reutersloader.dataset.transform = text.Tag()
elif "character" in transformer:
reutersloader.dataset.transform = text.Character()
elif "fw" in transformer:
reutersloader.dataset.transform = text.FunctionWord()
else:
print(u"No transformer set!")
exit()
# end if
# Dataset start
reutersloader.dataset.set_start(dataset_start)
# Set experience state
xp.set_state(space)
# Average sample
average_sample = np.array([])