"""Provide classes for dealing with Training Neural Networks. """ # standard modules import random class TrainingExample: """Hold inputs and outputs of a training example. XXX Do I really need this? """ def __init__(self, inputs, outputs, name = ""): self.name = name self.inputs = inputs self.outputs = outputs class ExampleManager: """Manage a grouping of Training Examples. This is meant to make it easy to split a bunch of training examples into three types of data: o Training Data -- These are the data used to do the actual training of the network. o Validation Data -- These data are used to validate the network while training. They provide an independent method to evaluate how the network is doing, and make sure the network gets trained independent of noise in the training data set. o Testing Data -- The data which are used to verify how well a network works. They should not be used at all in the training process, so they provide a completely independent method of testing how well a network performs. """ def __init__(self, training_percent = .4, validation_percent = .4): """Initialize the manager with the training examples. Arguments: o training_percent - The percentage of the training examples that should be used for training the network. o validation_percent - Percent of training examples for validating a network during training. Attributes: o train_examples - A randomly chosen set of examples for training purposes. o valdiation_examples - Randomly chosesn set of examples for use in validation of a network during training. o test_examples - Examples for training purposes. """ assert training_percent + validation_percent <= 1.0, \ "Training and validation percentages more than 100 percent" self.train_examples = [] self.validation_examples = [] self.test_examples = [] self.training_percent = training_percent self.validation_percent = validation_percent def add_examples(self, training_examples): """Add a set of training examples to the manager. Arguments: o training_examples - A list of TrainingExamples to manage. """ placement_rand = random.Random() # assign exact example randomly to the example types for example in training_examples: chance_num = placement_rand.random() # assign with the specified percentage if chance_num <= self.training_percent: self.train_examples.append(example) elif chance_num <= (self.training_percent + self.validation_percent): self.validation_examples.append(example) else: self.test_examples.append(example)