Simple Neural Network

A basic framework to construct a neural network system.

In this example,I used it to classify texts.(It is not well constructed)

Usage

Creating a dataset:

from dataset.dataset_fetcher import Dataset,Sinset
train_set = Dataset()#Classification dataset
sin_set = Sinset()#Regression dataset

Creating a network:

from nn.network import Network
network=Network(train_set)

Adding layers to the network:

network.append_linear_layer(node_count)
network.append_activation_layer(type="Sigmoid")#ReLU , Tanh , etc
......
network.append_linear_layer(output_node_count)

Training your network:

network.train_repeatly(train_count)

Reading result:

network.final_result

Network evaluation:

from util.evaluator import Evaluator:
#using train set as the test set to self-evaluate
self_evaluator = Evaluator(train_set)
self_evaluator.clf_evaluate()#classification evaluating
#self_evaluator.reg_evaluate()#regression evaluating

Dependencies

• Numpy
• Matplotlib
• Scikit learn(Used to fetch text training source)

To do

• Sine function is well predicted
• Text classifier did not work well,accuracy remains about 40%,suffering from overfitting problem.
• Mnist classifier works fine,accuray reaches 88%+