Data

The Data components specify different types of data to train the model with. See here for a list of supported data files. The following data components are available to import the data into the model:

• Local

• Environment

• Random

• IO Input

Local

Allows you to specify file(s) containing data to be used by the model. PerceptiLabs' Local Component currently supports the following file formats: .csv .txt .npy (NumPy file) .npz (NumPy zipped file). The component can also be configured to point to a directory of image files in the following formats: .jpg, .png, .jpeg, .tiff, and.tif. The files are read by PerceptiLabs in alphabetical/numerical order.

Parameters

Preview

Load data Choose files: selects the data to train. Choose folders: selects the location where data files are stored. Split dataset: partitions the dataset into the following groups: Training: core training data on which to train the model. Validate (aka verification data): data used to test model fit during training. Test: data to test the model against after training, to see how well the trained model handles data it hasn't seen before. Shuffle: randomizes the order to train the data on each epoch, to make the model more robust. Note that if you have multiple Data components which correspond to each other (e.g., one for images and one for the corresponding labels), you must ensure that Shuffle is configured the same for all of them.

Note: Paths cannot contain non-ascii characters.

Environment

Specifies that data be read in from a Gym environment. The Environment component is a collection of test problems — environments — that you can use to work out your reinforcement learning algorithms provided by OpenAI.

Note

This component requires installation dependencies before it can be used. See Setting up OpenAI Gym Environments for Reinforcement Learning for more information.

Parameters:

• Choose Engine: sets the environment to Gym or Unity.

• Environment: sets the pre-created Gym or Unity environment. The options available are based on the selected engine.

• Choose the .yaml file for your environment (Unity engine only): specifies the Unity environment YAML file to use.

Random

Generates random data (e.g., random noise). This can be useful for models such as GANs which are seeded with random noise to start with.

Parameters:

• Distribution: specifies the random distribution function to sample the values from. PerceptiLabs currently supports Normal (aka Gaussian) and Uniform. Depending on the distribution selected, the following parameters can be set:

• Normal parameters:

  • Mean: the mean value around which the random function will center the random values.

  • Stddev: the standard deviation that the random function will use. The larger the value, the more spread out the random values will be.

• Uniform parameters:

  • Min: specifies the lower bound of the distribution.

  • Max: specifies the upper bound of the distribution.

• Shape: the dimension and amount of random values to use.

IO Input

Allows you to specify a .csv file containing data to be used by the model.

Parameters:

• Load CSV data: selects a .csv file containing data on which to train.

• Split dataset: partitions the dataset into the following groups:

  • Training: core training data on which to train the model.

  • Validate (aka verification data): data used to test model fit during training.

  • Test: data to test the model against after training, to see how well the trained model handles data it hasn't seen before.

• Shuffle: randomizes the order to train the data on, to make the model more robust. Note that if you have multiple Data components which correspond to each other (e.g., one for images and one for the corresponding labels), you must ensure that Shuffle is configured the same for all of them.

• Feature name: name of the feature.