getting-started.py

# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "marimo",
#     "datasets",
#     "huggingface-hub",
# ]
# ///
"""
Getting Started with Hugging Face Datasets

This marimo notebook works in two modes:
- Interactive: uvx marimo edit --sandbox getting-started.py
- Script: uv run getting-started.py --dataset squad

Same file, two experiences.
"""

import marimo

app = marimo.App(width="medium")


@app.cell
def _():
    import marimo as mo
    return (mo,)


@app.cell
def _(mo):
    mo.md(
        """
        # Getting Started with Hugging Face Datasets

        This notebook shows how to load and explore datasets from the Hugging Face Hub.

        **Run this notebook:**
        - Interactive: `uvx marimo edit --sandbox getting-started.py`
        - As a script: `uv run getting-started.py --dataset squad`
        """
    )
    return


@app.cell
def _(mo):
    mo.md(
        """
        ## Step 1: Configure

        Choose which dataset to load. In interactive mode, use the controls below.
        In script mode, pass `--dataset` argument.
        """
    )
    return


@app.cell
def _(mo):
    import argparse

    # Parse CLI args (works in both modes)
    parser = argparse.ArgumentParser()
    parser.add_argument("--dataset", default="stanfordnlp/imdb")
    parser.add_argument("--split", default="train")
    parser.add_argument("--samples", type=int, default=5)
    args, _ = parser.parse_known_args()

    # Interactive controls (only shown in notebook mode)
    dataset_input = mo.ui.text(value=args.dataset, label="Dataset")
    split_input = mo.ui.dropdown(["train", "test", "validation"], value=args.split, label="Split")
    samples_input = mo.ui.slider(1, 20, value=args.samples, label="Samples")

    mo.hstack([dataset_input, split_input, samples_input])
    return args, argparse, dataset_input, parser, samples_input, split_input


@app.cell
def _(args, dataset_input, mo, samples_input, split_input):
    # Use interactive values if available, otherwise CLI args
    dataset_name = dataset_input.value or args.dataset
    split_name = split_input.value or args.split
    num_samples = samples_input.value or args.samples

    print(f"Dataset: {dataset_name}, Split: {split_name}, Samples: {num_samples}")
    return dataset_name, num_samples, split_name


@app.cell
def _(mo):
    mo.md(
        """
        ## Step 2: Load Dataset

        We use the `datasets` library to stream data directly from the Hub.
        No need to download the entire dataset first!
        """
    )
    return


@app.cell
def _(dataset_name, split_name):
    from datasets import load_dataset

    print(f"Loading {dataset_name}...")
    dataset = load_dataset(dataset_name, split=split_name)
    print(f"Loaded {len(dataset):,} rows")
    print(f"Features: {list(dataset.features.keys())}")
    return dataset, load_dataset


@app.cell
def _(mo):
    mo.md(
        """
        ## Step 3: Explore the Data

        Let's look at a few samples from the dataset.
        """
    )
    return


@app.cell
def _(dataset, mo, num_samples):
    # Select samples and display
    samples = dataset.select(range(min(num_samples, len(dataset))))
    df = samples.to_pandas()

    # Truncate long text for display
    for col in df.select_dtypes(include=["object"]).columns:
        df[col] = df[col].apply(lambda x: str(x)[:200] + "..." if len(str(x)) > 200 else x)

    print(df.to_string())  # Shows in script mode
    mo.ui.table(df)  # Shows in interactive mode
    return df, samples


@app.cell
def _(mo):
    mo.md(
        """
        ## Next Steps

        - Try different datasets: `squad`, `emotion`, `wikitext`
        - Run on HF Jobs: `hf jobs uv run --flavor cpu-basic ... getting-started.py`
        - Check out more UV scripts at [uv-scripts](https://huggingface.co/uv-scripts)
        """
    )
    return


if __name__ == "__main__":
    app.run()