What is Supervised Dataset in PyBrain?

In this article, we will be looking at the various uses and functionality of the supervised dataset in Pybrain.

A Dataset is a collection of data where we do give the list of values to each member belonging to the dataset. A supervised dataset following supervised learning has input and output fields. In this example, we will learn how to use a Supervised Dataset with PyBrain.To install Pybrain refer to How to Install PyBrain?

Let’s create a OR table where we have an input in form of a 2-D array and we get one output.

0 or 0 -> 0
0 or 1 -> 1
1 or 0 -> 1
1 or 1 -> 1

Libraries used:

• buildNetwork is a simple way to create networks that are composed of Modules connected with Connections.
• TanhLayer: After building a network we have to use some layer either TanhLayer or SoftmaxLayer. We will be using TanhLayer in our example.
• SupervisedDataSet: We have to set two values to input and target fields.
• BackpropTrainer: For training according to the supervised dataset

Example:

In this example, after building a network we will create two datasets one for training and another for testing.

Python

# importing buildNetwork
# from pybrain.tools.shortcuts
from pybrain.tools.shortcuts import buildNetwork

# importing TanhLayer
# from pybrain.structure
from pybrain.structure import TanhLayer

# importing SupervisedDataSet
# from pybrain.datasets
from pybrain.datasets import SupervisedDataSet

# importing BackpropTrainer
# from pybrain.trainers
from pybrain.supervised.trainers import BackpropTrainer

# creating a network with TanhLayer
# two input, two hidden and on output
network = buildNetwork(2, 2, 1, bias=True, hiddenclass=TanhLayer)

# Creating a dataset for training
# 2 output
# 1 input
or_train = SupervisedDataSet(2, 1)

# Creating a dataset for testing.
or_test = SupervisedDataSet(2, 1)

# Adding sample input
# 0 or 0 -> 0
# 0 or 1 -> 1
# 1 or 0 -> 1
# 1 or 1 -> 1
or_train.addSample((0, 0), (0,))
or_train.addSample((0, 1), (1,))
or_train.addSample((1, 0), (1,))
or_train.addSample((1, 1), (1,))

# Similarly adding samples for or_test
or_test.addSample((0, 0), (0,))
or_test.addSample((0, 1), (1,))
or_test.addSample((1, 0), (1,))
or_test.addSample((1, 1), (1,))

# Training network with dataset or_train.
trainer = BackpropTrainer(network, or_train)

# 1000 iteration on training data.
for iteration in range(1000):
  trainer.train()

# Testing data
trainer.testOnData(dataset=or_test, verbose = True)

Output: