Introducing Trieste

Trieste is an archival data format that facilitates documentation of data, data provenance tracking, and forensic analysis. If any of the following scenarios sound vaguely familiar, you may benefit from using Trieste:

Scenario I

While rushing to meet the deadline, you need to send some data to a colleague. You quickly make an ASCII file containing the requested data, using an interactive Python session (IPython). You don’t expect that you will need to create a similar table in the foreseeable future, and you are rushing, so you don’t take the time to write a script to automate the process. You simply send the file and forget about it. A few months later, you need to create a similar table from slightly different input data. You wish that you had at taken some notes, regarding how you produced the table so that you didn’t have to start over from scratch.

Scenario II

A colleague sends you a FITS file containing data that you would like to use. The file was created by your colleague’s friend, who received the input data from someone else, but he can’t remember who created the input files that were used in creating this file. You have been told that the file was originally part of a larger set of files, which included some documentation, but the documentation has evidently been lost. You spend a great deal of time inspecting the content of the FITS file in order to understand it. You would like to ask the file’s creator a question, but you can’t.

Scenario III

You need to produce thousands of multi-layer images of varying size and would like to put all of the images into a single file. You are frustrated because FITS doesn’t allow you to easily group sets of extensions. The layers of each image consist of different data types (64 bit floats and 32 bit integers) and the different layers have different metadata, so storing the images in multi-dimensional arrays within a FITS file won’t work. Furthermore, you don’t want to store thousands of FITS files in a .zip or .tar file because it is highly inconvenient for the recipient of the files to manually extract the data and then load it. This approach would also tax the file system because thousands of individual files would need to be created each time a new zip archive is loaded. You have heard that the HDF5 format could handle your use-case, but you would have to learn to create HDF5 files and explain to the recipient how to load the files. There must be a better way to do this!


Here are some examples to illustrate what Trieste is and what it does:

Installing Trieste

The easiest way to install Trieste is by using pip:

$ pip install trieste

Note that Trieste will only work with Python 3.6 or newer. If you have multiple Python installations, you may have to specify pip3.6:

$ pip3.6 install trieste

You can also download the source from GitHub.


Trieste files are intended to contain archival data. Once written, they are not intended to be modified—only read. They can store $N$-dimensional arrays, tables, and collections of arrays or tables. Every array, table, and collection has a name and contains a metadata dictionary (a map data structure) that can be augmented by the file’s author to provide documentation regarding the content of each object stored in the file, as well as instructions on using the data in the file. The goal is to provide the person reading the file with as much documentation and forensic information as possible about the file’s contents and the process by which the file was created.

The current prototype of Trieste (version 0.1.x) is a Python-specific format consisting of a specially-formatted, compressed NumPy .npz file in which every data object (NumPy array) has a corresponding metadata dictionary. Some of the metadata attributes are automatically added. Others, such as documentation strings (comments / READMEs) and object names, are strongly encouraged, but can be left empty. The automatically-added metadata are intended for software version compatibility checks and record-keeping / traceability. For example, the following are automatically added to the metadata:

In this module, there are 2 primary stand-alone functions:

  1. load(): for loading a Trieste file from the file system.

  2. save(): for saving a Trieste file to the file system.

There are 4 classes:

  1. Array: for storing $N$-dimensional arrays.

  2. Table: for storing Arrays with labeled columns of potentially different data types (analogous with a spreadsheet, ASCII table, or database table).

  3. Collection: for storing multiple Array or Table objects. Collections are only allowed to store a single type of object. For example, a collection can store multiple 2-D Array objects (like layers of an image) or multiple Table objects, but not a Table and an Array. Furthermore, the names of the objects in a Collection must be unique. This allows Collection objects to be indexed using the name of the object, so the syntax collection['red'] can be used to access the object in the collection whose name is red. Collection objects are also ordered containers, which means that they can be indexed by position, with an integer subscript, as in collection[3]. Collection objects are iterable, so that the syntax for object in collection: can be used to iterate through the contents.

  4. File: for interfacing with a file, after the file has been loaded.

    There are two types of files, in general:

    When loading a file containing only one object, the load function constructs the object itself (i.e., an instance of Array, Table, or Collection). When a file containing multiple objects is loaded, a File instance is created. Just like Collection objects, File objects are iterable and can be indexed by object name or position.

Future Plans

The 0.1.x series of Trieste is intended to test the basic ideas and the basic interface. It is a Python-specific file format which relies heavily upon the data serialization provided by NumPy, which uses Python’s Pickle.

If there is sufficient interest in the 0.1.x series, then a version 0.2.x branch will be opened to explore generalizing Trieste so that the file format can be written and read using a broad variety of programming languages (thus, making Trieste an interchange format as well as an archival format). Currently, it appears that MessagePack would be the best serialization format for this purpose.