Select Datastore for File Format or Application

A datastore is a repository for collections of data that are too large to fit in memory. Each file format and application uses a different type of datastore, which contains properties pertinent to the type of data or application that it supports. MATLAB^® provides datastores for standard file formats such as Excel^® files and datastores for specific applications such as Deep Learning. In addition to the existing datastores, if your data is in a proprietary format, then you can develop a customized datastore using the custom datastore framework.

Datastores for Standard File Formats

For a collection of data in standard file format use one of these options.

Datastore	Description
TabularTextDatastore	Text files containing column-oriented data, including CSV files
SpreadsheetDatastore	Spreadsheet files with a supported Excel format such as .xlsx
ImageDatastore	Image files, including formats that are supported by imread such as JPEG and PNG
ParquetDatastore	Parquet files containing column-oriented data
FileDatastore	Files with nonstandard file format Requires a custom file reading function

Transform or combine existing datastores.

Datastore	Description
CombinedDatastore	Datastore to combine data read from multiple underlying datastores
TransformedDatastore	Datastore to transform underlying datastore

Datastores to integrate with MapReduce and tall arrays.

Datastore	Description
KeyValueDatastore	Key-value pair data that are inputs to or outputs of mapreduce
TallDatastore	Datastore for checkpointing tall arrays

Datastores for audio and database data require additional toolboxes.

Datastore	Description	Toolbox Required
AudioDatastore	Datastore for collection of audio files	Audio Toolbox™
DatabaseDatastore	Datastore for collections of data in a relational database	Database Toolbox™

Datastores for Specific Applications

Based on your application use one of these datastores.

Application	Datastore	Description	Toolbox Required
Simulink Model Data	SimulationDatastore	Datastore for simulation input and output data that you use with a Simulink® model	Simulink
Simulation Ensemble and Predictive Maintenance Data	SimulationEnsembleDatastore	Datastore to manage simulation ensemble data	Predictive Maintenance Toolbox™
	FileEnsembleDatastore	Datastore to manage ensemble data in custom file format	Predictive Maintenance Toolbox
Measurement Data Format (MDF) Files	MDFDatastore	Datastore for collection of MDF files	Vehicle Network Toolbox™
	MDFDatastore	Datastore for collection of MDF files	Powertrain Blockset™
Deep Learning Datastores for preprocessing image or sequence data	PixelLabelDatastore	Datastore for pixel label data	Computer Vision Toolbox™ and Deep Learning Toolbox™
	PixelLabelImageDatastore	Datastore for training semantic segmentation networks Datastore is nondeterministic	Computer Vision Toolbox and Deep Learning Toolbox
	RandomPatchExtractionDatastore	Datastore for extracting random patches from images or pixel label images Datastore is nondeterministic	Image Processing Toolbox™ and Deep Learning Toolbox
	DenoisingImageDatastore	Datastore to train an image denoising deep neural network Datastore is nondeterministic	Image Processing Toolbox and Deep Learning Toolbox
	AugmentedImageDatastore	Datastore for resizing and augmenting training images Datastore is nondeterministic	Deep Learning Toolbox

Custom File Formats

For a collection of data in a custom file format, if each individual file fits in memory, use FileDatastore along with your custom file reading function. Otherwise, develop your own fully customized datastore for custom or proprietary data using the matlab.io.Datastore class. See Develop Custom Datastore.

Nondeterministic Datastores

Datastores that do not return the exact same data for a call to the read function after a call to the reset function are nondeterministic datastores. Do not use nondeterministic datastores with tall arrays, mapreduce, or any other code that requires reading the data more than once.

Some applications require data that is randomly augmented or transformed. For example, the AugmentedImageDatastore datastore, from the deep learning application augments training image data with randomized preprocessing operations to help prevent the network from overfitting and memorizing the exact details of the training images. The output of this datastore is different every time you perform a read operation after a call to reset.