jupyter labextension install @jupyterlab/dataregistry-extension
- Bring any data type you can imagine! Extensible and type safe data registry system.
- Register conversions between the different data types.
- Data changing on you? Use
RxJSobservables to represent data over time.
- Have a new way to look at your data? Create React or Phosphor components to view a certain type
- Built in data explorer UI to find and use available datasets.
- Dataset in your dataset? Use the nested datatype.
- Building another data centric application? Use the
@jupyterlab/dataregistrypackage which has no JupyterLab dependencies.
- Check out the project vision in the "Press Release from the Future"!
The data registry is a global collection of datasets. Each dataset is conceptually a tuple of
(URL, MimeType, cost, data), however, we store
them in nested maps of
Map<URL, Map<MimeType, [cost, data]>> so that for every unique pair of URL and MimeType we only have one dataset (
A "converter" takes in a dataset and returns several
other datasets that all have the same URL. We can apply a converter to a certain URL by viewing it as a graph exploration problem. There is one node per Mime Type and we can fill in the graph to add every reachable mime type with the lowest cost (
Conceptually, each Mime Type should correspond to some defined runtime type of data. For example
text/csv corresponds to an
Observable<string> which is the contents of CSV file. We need to be able to agree about these definitions so that if create
a converter to produce a
text/csv mime type and you create one that takes in that mime type and creates some visualization, we know we are dealing with the same type. A "data type" helps us here because we map a set of mime types to a TypeScript type. For example, we could define the CSV mime type as
new DataTypeNoArgs<Observable<string>>("text/csv"). We provide a way
to create a converter from one data type to another, which is
createConverter. Data types abstract away the textual representation of the mime type from the consumer of a data type and provide a type safe way to convert to or from that data type. All of our core conversions use this typed API (
void: Every URL starts with this data type when you ask for it. It has no actual data in it, so when you write a converter from it you will use the URL.
Observable<Set<URL_>>: This specifies the URLs that are "nested" under a URL. Use this if your dataset has some sense of children like a folder has a number of files in it or a database has a number of tables. These are exposed in the data explorer as the children in the hierarchy.
() => void: This is a function you can call to "view" that dataset in some way. It has a parameter as well, the "label", which is included in the mime type as an argument. This is exposed in the explorer as a button on the dataset.
I want to...
Explore my data in JupyterLab:
- Install JupyterLab >= 1.0
jupyter labextension install @jupyterlab/dataregistry
- Browse available datasets in the data explorer left side pane. We include support for viewing a few datasets. We plan on expanding this list and third party extension can extend it:
- Opening CSV files in the data grid and adding a snippet to open them with Pandas
- Opening PNG images in an image viewer
- Opening table data outputted in a notebook with
nteract's data explorer
Support a new data type or conversion:
You can either add support in this repo or by creating a new JupyterLab extension
that depends on the
IRegistry exposed by this extension. You can access a
Registry, which you can use to add your own converter.
It might also be useful to view the existing data types, by looking at the source code in this repo and by using the debugger. You can open this in JupyterLab by looking for the "Data Debugger" command:
Develop on this repo:
git clone https://github.com/jupyterlab/jupyterlab-data-explorer.git cd jupyterlab-data-explorer // (optional) Create a fresh conda environment // conda create -n jupyterlab-data-explorer -c conda-forge python=3.6 // conda activate jupyterlab-data-explorer // Install Jupyterlab pip install jupyterlab // Build and link the data explorer packages jlpm build:dev // Run Jupyterlab jupyter lab
This repo is in active development and we welcome any collaboration. If you have ideas or questions, feel free to open an issue. From there, we could setup a call to chat more in depth about how to work together. Please don't hesitate to reach out.
Or, feel free to tackle an existing issue or contribute a PR that you think improves things. We try to keep the current issues relevent and matched to relevent milestones, to give a sense on where this is going.
If the community grows around this we can adopt a more regular public meeting.