views/splash.py

import logging

import requests
import streamlit as st

from core.constants import OAUTH_CLIENT_ID
from core.past_projects import save_current_project
from core.path import get_resource_path
from core.query_params import set_project
from core.state import CurrentProject
from core.state import Metadata
import mlcroissant as mlc
from views.load import render_load
from views.previous_files import render_previous_files

_HUGGING_FACE_URL = "https://huggingface.co/datasets/"

_DATASETS = {
    "Titanic": ["data/embarkation_ports.csv", "data/genders.csv"],
    "FLORES-200": [],
    "GPT-3": [],
    "COCO2014": [],
    "PASS": [],
    "MovieLens": [],
    "Bigcode-The-Stack": [],
}

_INFO = """[Croissant](https://mlcommons.org/croissant) 🥐 is a high-level format for
machine learning datasets built
on [schema.org](https://schema.org/) and its Dataset vocabulary. A croissant
configuration file combines metadata, resource file descriptions, data structure, and
default ML semantics of dataset. You can familiarize yourself with the editor by
exploring the provided examples.

The editor supports creating a new configuration from scratch, as well as uploading
an existing Croissant JSON-MD file. Finally, you can also select any of your
past projects from the list.

You can change the project you are currently editing at any time by clicking
the Menu button and then choosing one of the options on this page."""


def render_splash():
    st.info(_INFO, icon="💡")
    if OAUTH_CLIENT_ID:
        st.info(
            "**Disclaimer**: Do not put sensitive information or datasets here. The"
            " storage on Hugging Face Spaces is ephemeral. If you want to host your own"
            " version locally, build the app from [the GitHub"
            " repository](https://github.com/mlcommons/croissant/tree/main/editor)."
        )
    col1, col2 = st.columns([1, 1], gap="large")
    with col1:
        with st.expander("**Create a new dataset**", expanded=True):

            def create_new_croissant():
                st.session_state[Metadata] = Metadata()
                save_current_project()

            st.button(
                "Create",
                on_click=create_new_croissant,
                type="primary",
            )
        with st.expander("**Load an existing dataset**", expanded=True):

            def create_example(dataset: str):
                base = f"https://raw.githubusercontent.com/mlcommons/croissant/main/datasets/{dataset.lower()}"
                url = f"{base}/metadata.json"
                try:
                    json = requests.get(url).json()
                    metadata = mlc.Metadata.from_json(mlc.Issues(), json, None)
                    st.session_state[Metadata] = Metadata.from_canonical(metadata)
                    save_current_project()
                    # Write supplementary files.
                    files = _DATASETS.get(dataset, [])
                    for file in files:
                        path = get_resource_path(file)
                        json = requests.get(f"{base}/{file}")
                        path.write_bytes(json.content)
                except Exception as exception:
                    logging.error(exception)
                    st.error(
                        "Sorry, it seems that the example is broken... Can you please"
                        " [open an issue on"
                        " GitHub](https://github.com/mlcommons/croissant/issues/new)?"
                    )

            dataset = st.selectbox(
                label="Canonical dataset",
                options=_DATASETS.keys(),
            )
            st.button(
                f"{dataset} dataset",
                on_click=create_example,
                type="primary",
                args=(dataset,),
            )
            url = st.text_input(
                label="Hugging Face dataset",
                placeholder="Example: https://huggingface.co/datasets/mnist",
            )
            if url.startswith(_HUGGING_FACE_URL):
                name = url.replace(_HUGGING_FACE_URL, "")
                api_url = (
                    f"https://datasets-server.huggingface.co/croissant?dataset={name}"
                )
                json = requests.get(api_url, headers=None).json()
                try:
                    metadata = mlc.Metadata.from_json(mlc.Issues(), json, None)
                    st.session_state[Metadata] = Metadata.from_canonical(metadata)
                    save_current_project()
                    st.rerun()
                except Exception:
                    st.error(f"Malformed JSON: {json}")
            elif url:
                st.error(
                    f"Unknown URL {url}. Hugging Face URLS should look like"
                    f" {_HUGGING_FACE_URL}somedataset."
                )
            render_load()
    with col2:
        with st.expander("**Recent projects**", expanded=True):
            render_previous_files()