With Lumipy, you can use the lumipy.provider submodule to create data providers that connect Python data science applications to Luminesce.

Prerequisites and setup


Before you begin creating Python providers, ensure you have setup your config with a Personal Access Token (PAT). For Windows, we recommend you run Windows PowerShell as administrator to send the following command, replacing <your-domain> and <your-access-token> with your own values:

$ lumipy config add --domain=<your-domain> --token=<your-access-token> 

Note: By default, providers are only visible to you. You can make a provider visible to all users in your domain by adding --user==global and --whitelist-me to the above command.

Send the following command to run a provider and complete your setup…

$ lumipy run demo

…you should see provider information followed by a browser window opening, prompting you to sign in. After signing in, you should see the following: 

Providers are ready to use.
Use ctrl+c or the stop button in jupyter to shut down

Building and running provider classes

Once you have completed the above setup, you can begin creating a Python provider. This consists of two steps:

Step 1: Building a provider class 
To build a provider class you must first import the submodule:

import lumipy.provider as lp

All provider classes must be subclasses of lumipy.provider.BaseProvider and implement the following methods:

  • __init__(): Declare metadata such as the column and parameter content of the provider. These are built from the corresponding metadata objects under lumipy.provider and supplied to super().__init__.
  • get_data(): Produce the data that is returned by the provider. Takes in a limit value, a filter representation and parameter values from the query being processed, returning a dataframe.

Note: If you are creating a provider using a Pandas dataframe, you can use Lumipy's built-in PandasProvider class instead; see how to do this.

Step 2: Running the provider 
To run a provider, you must run an instance of the provider class in ProviderManager. This starts up the provider, ready for querying in Luminesce. Note you can also use the command line interface (CLI) to run Python files containing provider objects, CSV files, and more; see how to do this


For example, you could build and run a provider for simulating a set of coin flips, with two columns Label and Result, and one parameter Probability with a default value of 0.5. To do this, you might:

  1. Import the lumipy.provider submodule and other required packages for this example.
  2. Create a provider class that inherits from the BaseProvider class. This class should define the following:
    • An __init__() method declaring the column and parameter metadata for the provider. This, alongside a name for the provider, is then supplied to super().__init__.
    • A get_data() method for getting and returning data from the provider. The context argument is used to specify information for query where clauses, limits and parameters. For this example:
      • The limit value for the number of rows to return is set to 100 if a query doesn't specify a limit.
      • Parameter values are retrieved from the parameters dictionary and must be within a specified range; an error is thrown in Python if the value is out of range and this is reported back in the progress log and query status.
      • The data - in this example coin flips - is generated and a dataframe is returned.
  3. Instantiate the provider object using the CoinFlips class. Finally, pass this into ProviderManager and use the run() method to start the provider.
# 1. Import submodule 
import lumipy.provider as lp
from pandas import DataFrame
from typing import Union, Iterator
import numpy as np

# 2. Create a provider class
class CoinFlips(lp.BaseProvider):
    def __init__(self):
        columns = [
            lp.ColumnMeta('Label', lp.DType.Text),
            lp.ColumnMeta('Result', lp.DType.Int),
        params = [lp.ParamMeta('Probability', lp.DType.Double, default_value=0.5)]
        # Supply above to super().__init__
        super().__init__('example.coin.flips', columns, params)
    def get_data(self, context) -> Union[DataFrame, Iterator[DataFrame]]:
        # If no limit is given, default to 100 rows.
        limit = context.limit()
        if limit is None:
            limit = 100
        # Get parameter value from params dict and throw an error if the parameter is out of given range 
        p = context.get('Probability')
        if not 0 <= p <= 1:
            raise ValueError(f'Probability must be between 0 and 1. Was {p}.')
        # Generate the coin flips and return. 
        return DataFrame({'Label':f'Flip {i}', 'Result': np.random.binomial(1, p)} for i in range(limit))
# 3. Pass into provider manager and run
coin_flips = CoinFlips()

Once running the provider, you can query it via Luminesce until you choose to shut it down:

You could choose to save the code above as a Python file, for example, and use the command line interface to run your Python provider on the fly:

$ lumipy run <path/to/>

You can run providers in this way for:

  • .py files containing provider objects
  • .csv files
  • Directories containing .csv and .py files

Building providers for Pandas dataframes

You can use the built-in PandasProvider class to easily pass a Pandas dataframe into the provider manager and make its data available in Luminesce.

To do this, simply input a dataframe object to the PandasProvider class and use the run() method with the provider manager. For example, to build a provider for one of the transaction source files from this tutorial, you might:

  1. Read data into Pandas from a CSV file.
  2. Pass the Pandas dataframe object and a provider name to the built-in PandasProvider class.
  3. Pass the provider object to the provider manager and use the run() method to start up the provider.
# 1. Read data from CSV
import pandas as pd
import lumipy.provider as lp
transactions_df = pd.read_csv("data/transactions.csv")

# 2. Pass into PandasProvider class
my_df = lp.PandasProvider(transactions_df, 'transactions')

# 3. Pass into provider manager and run

Once running, this provider will appear as pandas.transactions in Luminesce. You can query it until you choose to shut it down. Note you can set name_root = None in PandasProvider to change the provider name prefix.

If the input to PandasProvider is not a dataframe, the constructor passes the value into pandas.read_csv and uses the resultant dataframe. You can use this approach to build a provider from:

  • A local filepath
  • A URL
  • An IO stream
  • Anything else the function supports

For example, you could modify the code from steps 1-3 above to pass a CSV file into PandasProvider before running the provider manager:

import lumipy.provider as lp
file = "data/transactions.csv"

my_df = lp.PandasProvider(file, 'CreateDataframeForMe')

The provider runs as usual, allowing you to query the resultant dataframe via Luminesce:

Appendix A: .NET for Mac users

Mac users with an Apple silicon processor may experience the following error when attempting to run a provider locally:

Unable to load shared library 'SQLite.Interop.dll' or one of its dependencies.

You can follow these steps to overcome this issue:

  1. Install Rosetta 2. You can do this via the command line:
    $ softwareupdate --install-rosetta

    Once installed, send the following command to locate Rosetta:

    $ find / -name rosetta

    Add the resultant path to your $PATH environment variable. Read the Apple documentation on how to do this.

  2. Install the x86 version of .NET 6.0 or above. To do this, you can download the .NET 6.0 x64 installer for macOS and copy to a folder on your $PATH environment variable. Alternatively, you can send the following command:
    $ arch -x86_64 brew install dotnet-sdk

You can send the following command to check the file path is correct, containing a x64 folder, for example /usr/local/share/dotnet/x64/shared/Microsoft.AspNetCore.App:

$ dotnet --list-runtimes

If the path does not contain a x64 folder, uninstall .NET and reinstall the .NET x64 installer for macOS as mentioned in step 2 above. You should also ensure there are no .NET Arm64 versions on your path.