mlcore-client 0.1.3

A Python client for MLCore, providing a database-like interface for managing datasets and machine learning models.

MLCore Python Client

A powerful Python client for MLCore, providing a database-like interface for managing datasets and machine learning models. Built for data scientists and engineers who want to automate their machine learning workflows.

Links

• GitHub Repository: Amanbig/MLCore
• Docker Hub: procoder588/mlcore

Installation

pip install mlcore-client

Connection

Connect to your MLCore instance using a connection string (URI) or explicit parameters.

from mlcore import MLCore

# Option 1: Connection string (Recommended)
# Format: mlcore://user:password@host:port
client = MLCore("mlcore://admin:password@localhost:8000")

# Option 2: Explicit parameters
client = MLCore(
    host="localhost", 
    port=8000, 
    email="[e-mail]", 
    password="password"
)

Asynchronous Support

For asyncio applications, use MLCoreAsync:

from mlcore import MLCoreAsync

client = MLCoreAsync("mlcore://admin:password@localhost:8000")
await client.connect()
# ... use await with all methods ...
await client.close()

---

API Documentation

📊 Dataset Manager (client.datasets)

Manage data lifecycle from raw files to cleaned versions.

Method	Description	Parameters
list()	List all accessible datasets.	-
get(id)	Get metadata for a dataset.	id: UUID or string
upload_file(path)	Upload a raw CSV/Excel file.	path: Local file path
create(...)	Register a file as a dataset.	name, description, file_id, rows, columns, metadata
get_data(id, ...)	Fetch paginated rows.	id, page=1, limit=50, as_df=False
clean(id, ...)	Apply cleaning logic.	id, strategy ('drop_nulls', 'fill_mean', etc.), columns
transform(id, ...)	Apply transformations.	id, strategy ('standard_scaler', etc.), columns
get_versions(id)	Get history/lineage.	id
delete(id)	Permanently remove dataset.	id

🤖 Model Manager (client.models)

Train, evaluate, and deploy machine learning models.

Method	Description	Parameters
list()	List all trained models.	-
get(id)	Get model specs & metrics.	id: UUID or string
train(...)	Start a training job.	dataset_id, algorithm, target_column, features, hyperparameters, name
predict(id, inputs)	Run real-time inference.	id, inputs: Dict[feature_name, value]
download(id, path)	Download .joblib file.	id, path: Local destination path
retrain(id, ...)	Run new training session.	id, dataset_id, algorithm, etc.
get_hyperparameters(algo)	Get valid hyperparams.	algo: Algorithm name
get_versions(id)	Get model history.	id
update_meta(id, ...)	Rename/update description.	id, name, description
delete(id)	Delete model artifacts.	id

📈 General Methods

Method	Description
get_stats()	Get platform-wide statistics (counts of models, datasets, files).
health_check()	Check server connectivity and version.

---

Usage Examples

End-to-End Pipeline

# 1. Prepare Data
file_info = client.datasets.upload_file("raw_data.csv")
ds = client.datasets.create(name="Training Set", file_id=file_info["id"], ...)
cleaned_ds = client.datasets.clean(ds["id"], strategy="drop_nulls")

# 2. Train Model
model = client.models.train(
    dataset_id=cleaned_ds["id"],
    algorithm="random_forest",
    target_column="label",
    hyperparameters={"n_estimators": 200}
)

# 3. Monitor Specs
print(f"Model trained with {model['accuracy']}% accuracy")

# 4. Predict
result = client.models.predict(model["id"], inputs={"feature_1": 0.5, "feature_2": 1.2})
print(f"Prediction: {result['predictions']}")

Features

• Database-like Connection: URI-based connection strings (mlcore://).
• Session Persistence: Automatic token management and re-auth logic.
• Pandas Ready: One-click conversion from server data to DataFrames.
• Async First: First-class support for httpx-based asynchronous I/O.
• Developer Friendly: Strictly typed and linted with Ruff.

License

MIT