def main():
import numpy as np
import lxmls.readers.sentiment_reader as srs
from lxmls.deep_learning.utils import AmazonData
corpus = srs.SentimentCorpus("books")
data = AmazonData(corpus=corpus)
# Model
geometry = [corpus.nr_features, 20, 2]
activation_functions = ['sigmoid', 'softmax']
# Optimization
learning_rate = 0.05
num_epochs = 10
#batch_size = 30
batch_size = 1
model = NumpyMLP(geometry=geometry,
activation_functions=activation_functions,
learning_rate=learning_rate)
from lxmls.deep_learning.mlp import get_mlp_parameter_handlers, get_mlp_loss_range
# Get functions to get and set values of a particular weight of the model
get_parameter, set_parameter = get_mlp_parameter_handlers(layer_index=1,
is_bias=False,
row=0,
column=0)
# Get batch of data
batch = data.batches('train', batch_size=batch_size)[0]
# Get loss and weight value
current_loss = model.cross_entropy_loss(batch['input'], batch['output'])
current_weight = get_parameter(model.parameters)
# Get range of values of the weight and loss around current parameters values
weight_range, loss_range = get_mlp_loss_range(model, get_parameter,
set_parameter, batch)
# Get the gradient value for that weight
gradients = model.backpropagation(batch['input'], batch['output'])
print(gradients)
current_gradient = get_parameter(gradients)
params_svm_sd = svm.train(sd.train_X, sd.train_y)
y_pred_train = svm.test(sd.train_X, params_svm_sd)
acc_train = svm.evaluate(sd.train_y, y_pred_train)
y_pred_test = svm.test(sd.test_X, params_svm_sd)
acc_test = svm.evaluate(sd.test_y, y_pred_test)
fig, axis = sd.add_line(fig, axis, params_svm_sd, "SVM", "yellow")
print "SVM Online Simple Dataset Accuracy train: %f test: %f" % (acc_train,
acc_test)
print
# End of exercise 3.1 #########
# Exercise 3.2: implement Naive Bayes for multinomial data ########
# Read the book review data
scr = srs.SentimentCorpus("books")
# Initialize the Naive Bayes classifier for multinomial data
mnb = mnb.MultinomialNaiveBayes()
# Learn the NB parameters from the train data
params_nb_sc = mnb.train(scr.train_X, scr.train_y)
# Use the learned parameters to predict labels for the training data
y_pred_train = mnb.test(scr.train_X, params_nb_sc)
# Compute accuracy on training data from predicted labels and true labels
acc_train = mnb.evaluate(scr.train_y, y_pred_train)
# Use the learned parameters to predict labels for the test data
y_pred_test = mnb.test(scr.test_X, params_nb_sc)
def corpus():
return srs.SentimentCorpus("books")
def scr():
return srs.SentimentCorpus("books")
