Advanced uses: Webserver¶

This section assumes you already have a hiplot webserver running (otherwise, see Option 2: Use HiPlot webserver)

Experiments URI¶

In HiPlot server, experiments are loaded by entering a string in the textarea. It can be the path to a CSV file, or really anything that can uniquely specify which experiment we are trying to load. This string is the Experiment Universal Resource Identifier (or in short Experiment URI). HiPlot translates those URIs into hiplot.Experiment using experiment fetchers. We will see later how to write our own one (Make HiPlot server render your own experiments).

Compare multiple experiments¶

Multiple experiments can be combined together in the interface using the special multi fetcher. It allows 2 syntaxes:

Dictionary of named experiments¶

multi://{
    "exp1_name": "exp1_uri",
    "exp2_name": "exp2_uri"
}

List of experiments¶

multi://[
    "exp1_uri",
    "exp2_uri"
]

Make HiPlot server render your own experiments¶

About experiment fetchers¶

When we request an experiment in HiPlot, the server will call the experiment fetchers it has iteratively. Each fetcher can either:

Return a hiplot.Experiment, that will be sent to the client
Raise an hiplot.ExperimentFetcherDoesntApply exception, in which case the server moves on and tries the next fetcher

In order to avoid conflicts, it is good practice to use a prefix to determine which fetcher we want to call. Here we use myxp://

How we will do that¶

We are going to write our own experiment fetcher, to translate custom Experiments URI into hiplot.Experiment. We will do that in several steps:

First we will write a file my_fetcher.py that contains our fetcher function fetch_my_experiment
Then, we will restart HiPlot server with this additional fetcher: hiplot my_fetcher.fetch_my_experiment

Step 1: Create an experiment fetcher¶

An experiment fetcher transforms a string (that the user enters, a path for instance) into a proper hiplot.Experiment. Let’s write a dummy one that takes a folder, and returns the content of data.csv inside if the file exists.

# my_fetcher.py
from pathlib import Path
import hiplot as hip
def fetch_my_experiment(uri):
    # Only apply this fetcher if the URI starts with myxp://
    PREFIX="myxp://"
    if not uri.startswith(PREFIX):
        # Let other fetchers handle this one
        raise hip.ExperimentFetcherDoesntApply()
    uri = uri[len(PREFIX):]  # Remove the prefix

    return hip.Experiment.from_csv(uri + '/data.csv')

Step 2: Run HiPlot server with the new fetcher¶

Our fetcher is ready, let’s simulate a dummy experiment that we can load later

# Some dummy data for the demo
mkdir xp_folder
echo -e "col1, col2, col3\n1,2,3\n2,2,3\n4,4,2" > xp_folder/data.csv

>>> hiplot my_fetcher.fetch_my_experiment

In the interface, you can load the string myxp://xp_folder

Dump your experiments to CSV or HTML with `hiplot-render`¶

HiPlot also provides a script to generate CSV or a standalone HTML file for an experiment from the command line. Like the webserver, it can take additional fetchers to render custom experiments (see also Make HiPlot server render your own experiments above). You can get started with:

>>> hiplot-render --help

Table Of Contents