5.1 Create a step

A step is an atomic component defined by its input and output parameters and by the processing it applies. Steps are the building blocks of Pipelines. In practice, a step is a function with inputs and outputs coded in Python. They are assembled with each other to create a complete ML pipeline. The Python code (only available language for the moment) used by the step is stored on a GitHub repository.

An input of a step is an object you can use inside the code.

An output of a step is defined from the results of the step function.

You will be able to connect inputs & outputs of a step with another step to compose a complete ML pipeline by using a directed acyclic graph (DAG).

Each step is considered as a specific container that is executed on Kubernetes.

The steps are stored in a specific environment, and only people with access to this environment can read and write the steps.

Function name	Method	Return type	Description
Input	Input(name, data_type=”string”, description=””, is_required=False, default_value=None)	Input SDK Object	Create an Input object to give at create_steps() function for step a step input.
Output	Output(name, data_type=”string”, description=””)	Output SDK Object	Create an Output object to give at create_steps() function for step a step output.
create_step	create_step(step_name, function_path, function_name, repository_branch=None, description=None, container_config=None, inputs=None, outputs=None)	list of dict[str, str]	Create pipeline steps from a source code located on a remote repository.

5.1.2 Prepare your code on GitHub

Prerequisites: Before the creation of your first step, make sure you have already done this :

Setup Project & Environment
GitHub repository link to the project

❓ Why do you need to put your code on your GitHub repository ? This simplifies the access to the source code by the Craft AI platform. Indeed, the platform will be able to directly fetch your code from the repository, without the need for you to send it directly each time you change it, you just have to push it to your GitHub repository.

Currently, you can create a step via the Python SDK and not with graphical interface. But, after the creation, you will be able to see the step on the UI platform.

If it’s not already done, put the code of the step into a GitHub repository linked to the platform. The file with the entry function of your step can be anywhere in your GitHub repository.

Example tree file in repo :

| requirements.txt
| src
   | my_entry_function_step.py
...

Example my_entry_function_step.py :

import numpy as np
# and other import

def entryStep(dataX_input, dataY_input) :

    # Some machine learning code

    return result_output

5.1.3 Define step inputs and outputs

A step may need to receive some information or give some result (just like a function). To do that, we use Input and Output object. These objects allow defining the properties of the input or output that will be expected in the step. The input and output objects thus created must be given as a parameter of the step creation. Each input is defined as an Input object and, each Output is defined as an Output object, through a class available in the SDK.

5.1.3.1 Input object definition

from craft_ai_sdk.io import Input

Input(
   name="*your_input_name*",
   data_type="*your_io_data_type*",
   description="",
   is_required=True
   default_value="*default_value*"
)

Parameters

name just a name for identifying the input later.
data_type, one of the following possible types:
- file: reference to binary data, equivalent to a file’s content. If the input/output is not available, an empty stream.
- json: JSON-serializable Python object. The following sub-types are provided for more precise type checking, but they are all JSON
- string
- number
- array of JSON
- boolean
  
  If the input/output is not available, None in Python
default_value (optional) - If the parameter is empty, this value will be set by default. If a deployment receives an empty parameter and already put a default value in the input, the default value of deployment will be keep.
is_required (optional, True by default) - Push an error is the input is empty.
description (optional) - This parameter precise what it’s expected in this input. It’s not read by the machine, it’s like a comment.

Return

No return

5.1.3.2 Output object definition

from craft_ai_sdk.io import Output

Output(
   name="*your_input_name*",
   data_type="*your_io_data_type*",
   description="",
)

Parameters

name just a name for identifying the input later.
data_type, one of the following possible types:
- file: reference to binary data, equivalent to a file’s content. If the input/output is not available, an empty stream.
- json: JSON-serializable Python object. The following sub-types are provided for more precise type checking, but they are all JSON
- string
- number
- array of JSON
- boolean
If the input/output is not available, None in Python
description (optional) - This parameter precise what it’s expected in this input. It’s not read by the machine, it’s like a comment.

Return

No return

Note

You can use craft_ai_sdk.INPUT_OUTPUT_TYPES to get all possible types in Input and Output objects.

List of all possible types :

ARRAY = “array”
BOOLEAN = “boolean”
FILE = “file”
JSON = “json”
NUMBER = “number”
STRING = “string”

Example :

from craft_ai_sdk.io import Input, INPUT_OUTPUT_TYPES

Input(
   name="inputName",
   data_type=INPUT_OUTPUT_TYPES.JSON,
)

5.1.3.3 Example for input and output

Input(
    name="inputName",
    data_type="string",
    description="A parameter for step input",
    is_required=True,
    default_value="default_content_here"
)

Output(
    name="inputName",
    data_type="string",
    description="A parameter for step input",
)

5.1.4 Create a step

5.1.4.1 Function definition

Create pipeline steps from a source code located on a remote repository.

sdk.create_step(
    function_path="src/my_reusable_funtion.py",
    function_name="my_function",
    inputs=[Input(...)],
    outputs=[Output(...)],
    [name="*your-custom-step-name*"], # by default its the function name
    [description="*text with limit*"],
    [repository_branch="*your-git-branch* or *your-git-tag*"],
    [container_config = {
        [language="python:3.8-slim"],
        [repository_url="*your-git-url*"],
        [repository_deploy_key="*your-private_key*"],
        [requirements_path="*your-path-to-requirements*"],
        [included_folders=["*your-list-of-path-to-sources*"]],
        [system_dependencies=["package_1", "package_2"]],
        [dockerfile_path="*your-dockerfile-path*"],
}],
)

Parameters

function_path (str) – Path to access to the file who had the entry function of the step.
function_name (str) – Function name of entry function step.
inputs (list<Input>) – List of step inputs.
outputs (list<Output>) – List of step outputs.
name (str) – Step name. By default, it’s the function name. The name must be unique inside an environment and without special character ( - _ & / ? …)
description (str, optional) – Description of the step, it’s no use by the code, it’s only for user.
repository_branch (str, optional) – Branch name for GitHub repository. Defaults to None.
container_config (dict, optional) – Dict Python object where each key can override default parameter values for this step defined at project level.
- language (str, optional) – Language of programming used inside the step. Defaults to “python:3.8-slim”. Versions “python:3.9-slim” and “python:3.10-slim” are also available.
- repository_url (str, optional) – Remote repository URL.
- repository_deploy_key (str, optional) – Private SSH key related to the repository.
- requirements_path (str, optional) – Path to the file requirement for Python dependency.
- included_folders (list, optional) – List of folders that need to be accessible from step code.
- system_dependencies (list, optional) – List of APT Linux packages to install.
- dockerfile_path (str, optional) – Path to a docker-file for having a custom config in step. (see the part after for more detail)

Note

The repository_branch parameters as well as the container_config elements (except dockerfile_path) can take one of the STEP_PARAMETER object’s values in addition to theirs.

In fact, STEP_PARAMETER allows us to specify at the step level whether we want to take the project’s values (default behavior) or define a null value:

STEP_PARAMETER.FALLBACK_PROJECT : Allows to take the value defined in the project parameters (default behavior if the field is not defined).
STEP_PARAMETER.NULL : Allows to set the field to null value and not to take the value defined in the project.

Example with a code step that does not need a requirement.txt and does not take the one defined in the project settings:

from craft_ai_sdk import STEP_PARAMETER

# Code for init SDK here ...

sdk.create_step(
  function_path="src/helloWorld.py",
  function_name="helloWorld",
  step_name="stepName",
  container_config = {
      "requirements_path": STEP_PARAMETER.NULL,
   }
)

Returns

List of steps represented as dict (with either key “name” if the creation succeeded or keys “name” and “error” if the creation failed). The return type is a list of dict[str, str].

ℹ️ Once our step is created, we need to create the pipeline that wraps the step. It is mandatory to create a pipeline once the step is created to be able to use it later. This technical choice was made in anticipation of future multistep functionality. This forces the use of a pipeline to contain the steps.

5.1.4.2 Example: Create step from scratch

Function usage

from craft_ai_sdk import Input, Output

input1 = Input(
    name="input1",
    data_type="string",
    description="A parameter named input1, its type is a string",
    is_required=True,
)

input2 = Input(
    name="input2",
    data_type="file",
    description="A parameter named input2, its type is a file"
)

input3 = Input(
    name="input3",
    data_type="number",
)

prediction_output = Output(
    name="prediction",
    data_type="file",
    default_value="default,content,here",
)

step = sdk.create_step(
        function_path="src/my_reusable_funtion.py",
        function_name="my_function",
    inputs_list=[input1, input2, input3],
        outputs_list=[prediction_output],
    description="Apply the model to the sea",
        ## ...
)

Returned object

{
    "name": "apply_model",
    "inputs_list": [
        {
            "name": "input1",
            "data_type": "string",
            "is_required": True,
            "description": ...,
        },
        ...
    ],
    "ouputs_list": [
        ...
    ],
    ...
}

Note

If you need to create a step with more specific configuration, you can do this with a custom dockerfile, more detail about here.