API Request

API Request Example

This example demonstrates a simple request to the mloda API. You describe WHAT data you need - mloda resolves HOW to get it.

Tip: For AI agents or quick prototyping, you can also use inline data with api_data - see the 30-second example.

In this example, mloda automatically determines that a CsvReader feature group will fulfill the request and respond with the resulting DataFrame.

1. Import the Required Modules

We first need to import the necessary components to set up our request:

from mloda.user import mloda
from mloda.user import DataAccessCollection

2. Define Data Sources

Next, we define the data source by specifying the file path and creating a DataAccessCollection object:

file_path = "tests/test_plugins/feature_group/src/dataset/creditcard_2023_short.csv"
data_access_collection = DataAccessCollection(files={file_path})

3. Define the features

We now specify the list of features we want to retrieve. The features are provided as a comma-separated string, which we split into a list:

feature_list = "id,V1,V2,V3,V4,V5,V6,V7,V8,V9,V10,V11,V12,V13,V14,V15,V16,V17,V18,V19,V20,V21,V22,V23,V24,V25,V26,V27,V28,Amount,Class"
feature_list = feature_list.split(",")

4. Execute the Request and Retrieve Results

Finally, we send the request to the mloda API, specifying the compute framework and the data source. The result is returned as a DataFrame:

result = mloda.run_all(
    feature_list,
    compute_frameworks=["PyArrowTable"],
    data_access_collection=data_access_collection)
result[0]

Expected output:

pyarrow.Table
V8: double
id: int64
...
V8: [[-0.13000604758867731,-0.13311827417649086,-0....]]
id: [[0,1,2,...]]
....

5. Summary

This example shows how easy it is to interact with the mloda API for data retrieval. By defining the data source and specifying the features, mloda handles the rest.

For Realtime / Inference Use Cases

With run_all(), mloda rebuilds the full execution plan on every call. That is fine for batch jobs, but in latency-sensitive scenarios the repeated planning overhead matters.

Typical examples where this applies:

ML model serving — a prediction endpoint computes the same derived features for every incoming request; only the raw input row changes.
Streaming / event-driven pipelines — each incoming event needs the same feature transformations applied in real time.
Interactive applications — a dashboard or API recalculates features on every user action with fresh parameters.

The two-phase API lets you pay the planning cost once at startup and then execute cheaply per request:

# 1. Prepare once (e.g. at server startup)
session = mloda.prepare(feature_list, compute_frameworks=["PyArrowTable"], data_access_collection=data_access_collection)

# 2. Execute per request with fresh data
result = session.run(api_data={"MyKey": {"col": [1, 2]}})

See mloda API — Two-Phase Execution for details.

For AI Agents: JSON-Based Requests

LLMs can generate JSON feature requests without writing Python code:

from mloda.user import load_features_from_config

llm_request = '["id", "V1", "V2", "Amount"]'
features = load_features_from_config(llm_request, format="json")
result = mloda.run_all(features=features, compute_frameworks=["PandasDataFrame"], ...)

See Feature Configuration for more details on JSON-based configuration.