Table of Contents

• TensorFlow Models
  • Purpose
  • Features
• Usage
  • NeuralNet
    • Initialization
    • Building the Model
    • L2 Loss for Objective Function
    • Reset Weights

TensorFlow Models

tensorflow-models - TensorFlow machine learning models.

Purpose

A collection of helpful models for streamlining the creation of machine learning models Developed in the CAMEL lab at Clarkson University for anomaly detection.

Features

Currently, just contains the NeuralNet model. Planning to add more models.

Usage

NeuralNet

A model for creating a fully connected feed forward neural net given a list of layer sizes and activation functions.

Initialization

We can create a model by passing a list of layer sizes and a list of activation functions.

model = models.NeuralNet.NeuralNet(
	[num_input, 100, num_output],
	[tf.nn.relu, tf.nn.sigmoid]
)

Building the Model

We can build the computation graph using the function create_network. An example is provided below

X = tf.placeholder('float', [None, num_inputs])
keep_prob = tf.placeholder('float')
prediction = model.create_network(X, keep_prob)

L2 Loss for Objective Function

The L2 loss can be added to the objective function by calling get_l2_loss.

cost += 0.01 * model.get_l2_loss()

Reset Weights

Using reset_weights returns a tensor operation to reset the weights using sess.run

with tf.Session() as sess:
	sess.run(model.reset_weights())