Component: Recipes

Description

Dataiku DSS comes with some existing Visual recipes allowing the user to perform standard analytic transformations. Visual recipes perform a transformation in the Flow, but they are not the only way to do it.

Coders can use Recipes based on code to code additional transformations particularly when Dataiku DSS does not cover their usage. Code recipes are a good way to extend Dataiku DSS’s capabilities. These user-defined code recipes can be converted to plugin recipes, allowing them to be distributed to non-coder users as a visual recipe.

To be able to convert a code recipe to a plugin recipe:

• Open the code recipe;

• Click on the “Action” button (or open the right panel);

• Click on the “Convert to plugin” button;

• Fill out the form.

There are two different ways to integrate a plugin recipe. You can choose to add the recipe plugin to an existing plugin. Or, you can choose to create a new plugin, and then add the plugin recipe to this new plugin.

A plugin recipe is configurable in the associated JSON file (automatically created by Dataiku DSS during the conversion process), in the folder custom-recipes/{<plugin-recipe-id>}/recipe.json. This JSON configuration file comprise different parts as shown in the code below.

{
    // Metadata section
    "meta": {
        //metadata used for display purpose
    },

    // Kind of the code recipe
    "kind": "PYTHON",

    // Input/Output section
    "inputRoles": [
        // DSS objects that can be used as input for the recipe
        // A recipe can have many different input objects
    ],
    "outputRoles": [
        // DSS objects that can be used as output for the recipe
        // A recipe can have many different output objects
    ],

    // Parameters section
    "params": [
        // Parameter definition usable for your recipe
    ],

    // Advanced configuration section
    // various configuration options.

}

Write the plugin recipe code in the file recipe.py. It should, at least, read the input parameter and produce the dataset mentioned in the output section, like described below.

# To retrieve the datasets of an input role named 'input_A' as an array of dataset names:
dataset_name = get_input_names_for_role('dataset_input')[0]

# For outputs, the process is the same:
output_name = get_output_names_for_role('dataset_output')[0]

# Read recipe inputs
df = dataiku.Dataset(dataset_name).get_dataframe()

# In this example, we just copy the input to the output

# Write recipe outputs
output_dataset = dataiku.Dataset(output_name)
output_dataset.write_with_schema(df)

Metadata

Metadata is used for display purposes. You can configure the name of the recipe as well as its description, the icon used to represent the recipe and also the color of this icon, by filling out the "meta" field as shown below.

Example

"meta": {
    // label: name of the recipe as displayed, should be short
    "label": "Short title",

    // description: longer string to help end users understand what this recipe does
    "description": "A longer description that helps the user understand the purpose of the recipe",

    // icon: must be one of the FontAwesome 3.2.1 icons, complete list here at https://fontawesome.com/v3.2.1/icons/
    "icon": "icon-thumbs-up-alt",

    // DSS currently supports the following colors: red, pink, purple, blue, green, sky, yellow, orange, brown, and gray.
    "iconColor": "blue"
},

Input/Output

Each input/output role has the following structure:

• name: Name of the role; this is how the role is referenced elsewhere in the code

• label: A displayed name for this role

• description: A description of what the role means

• arity: UNARY or NARY (can accept one or multiple values?)

• required (boolean, default false): Does this role need to be filled?

• acceptsDataset (boolean, default true): Whether a dataset can be used for this role

• acceptsManagedFolder (boolean, default false): Whether a managed folder can be used for this role

• acceptsSavedModel (boolean, default false): Whether a saved model can be used for this role

Grouping input/output into roles enables the coder (and the user) to group their input/output into semantic groups of data, rather than having all the input/output at the same level.

Example

"inputRoles": [
    {
        "name": "dataset_names",
        "label": "input A displayed name",
        "description": "what input A means",
        "arity": "NARY",
        "required": true,
        "acceptsDataset": true
    },
    {
        "name": "managed_folder_name",
        "label": "input B displayed name",
        "description": "what input B means",
        "arity": "UNARY",
        "required": false,
        "acceptsDataset": false,
        "acceptsManagedFolder": true
    }

Parameters

From a parameter perspective, a plugin code recipe is not different than other plugin components. However, for the plugin code recipe, there are two types of parameters (COLUMN and COLUMNS). They allow the user to select one (or more) column(s) in a dataset as parameter(s) for the plugin code recipe. For more about parameters, please see Parameters.

Advanced configuration

Select from Flow view

To make a new recipe directly available from the Flow view when selecting:

• a Dataset, add the field: "selectableFromDataset"

• a Managed Folder, add the field: "selectableFromFolder"

• a Saved Model, add the field: "selectableFromSavedModel"

Each added field should target an existing inputRole, see below for the usage.

Complete example

{
    "meta": {
        "label": "Clip values",
        "description": "Allow clipping values on different columns",
        "icon": "icon-align-justify",
        "iconColor": "orange"
    },
    "kind": "PYTHON",
    "inputRoles": [
        {
            "name": "dataset_name",
            "label": "Datasets to clip",
            "description": "Automatically remove outliers from a dataset, by clipping the values",
            "arity": "UNARY",
            "required": true,
            "acceptsDataset": true
        },
        {
            "name": "managed_folder_name",
            "label": "dummy input",
            "description": "just here to demonstrate selection from the flow",
            "arity": "UNARY",
            "required": false,
            "acceptsDataset": false,
            "acceptsManagedFolder": true
        }
    ],
    "outputRoles": [
        {
            "name": "dataset_clipped",
            "label": "Clipped dataset",
            "description": "Result of the clipping",
            "arity": "UNARY",
            "required": true,
            "acceptsDataset": true
        }
    ],
    "params": [
        {
            "name": "IQR_coeff",
            "type": "DOUBLE",
            "defaultValue": 1.5,
            "label": "IQR coefficient",
            "mandatory": true
        }
    ],
    "selectableFromDataset": "dataset_name",
    //The next selectable is just for the example
    "selectableFromFolder": "managed_folder_name"

    // The field "resourceKeys" holds a list of keys that allows limiting the number
    // of concurrent executions and activities triggered by this recipe.
    //
    // Administrators can configure the limit per resource key in the Administration > Settings > Flow build
    // screen.
    "resourceKeys": []
}