demo_list.py

import dataclasses
import datetime
import operator
import pathlib

import numpy as np
import pandas as pd
import tqdm.auto
import yaml
from huggingface_hub import HfApi

from constants import SLEEP_TIME_INT_TO_STR, SLEEP_TIME_STR_TO_INT


@dataclasses.dataclass(frozen=True)
class DemoInfo:
    space_id: str
    url: str
    title: str
    owner: str
    sdk: str
    sdk_version: str
    likes: int
    status: str
    last_modified: str
    sleep_time: int
    replicas: int
    private: bool
    hardware: str
    suggested_hardware: str
    created: str = ""

    def __post_init__(self) -> None:
        object.__setattr__(self, "last_modified", DemoInfo.convert_timestamp(self.last_modified))
        object.__setattr__(self, "created", DemoInfo.convert_timestamp(self.created))

    @staticmethod
    def convert_timestamp(timestamp: str | datetime.datetime) -> str:
        if isinstance(timestamp, datetime.datetime):
            return timestamp.strftime("%Y/%m/%d %H:%M:%S")
        try:
            dt = datetime.datetime.strptime(timestamp, "%Y-%m-%dT%H:%M:%S.%fZ").astimezone(datetime.timezone.utc)
            return dt.strftime("%Y/%m/%d %H:%M:%S")
        except ValueError:
            return timestamp

    @classmethod
    def from_space_id(cls, space_id: str) -> "DemoInfo":
        api = HfApi()
        space_info = api.space_info(repo_id=space_id)
        card = space_info.cardData
        runtime = space_info.runtime

        return cls(
            space_id=space_id,
            url=f"https://huggingface.co/spaces/{space_id}",
            title=card.get("title", ""),
            owner=space_id.split("/")[0],
            sdk=card["sdk"],
            sdk_version=card.get("sdk_version", ""),
            likes=space_info.likes,
            status=runtime.stage,
            last_modified=space_info.lastModified,
            sleep_time=runtime.sleep_time or 0,
            replicas=runtime.raw["replicas"]["current"] or runtime.raw["replicas"]["requested"],
            private=space_info.private,
            hardware=runtime.hardware or runtime.requested_hardware or "",
            suggested_hardware=card.get("suggested_hardware", ""),
            created=space_info.created_at,
        )


def get_df_from_yaml(path: pathlib.Path | str) -> pd.DataFrame:
    with pathlib.Path(path).open() as f:
        data = yaml.safe_load(f)
    demo_info = []
    for space_id in tqdm.auto.tqdm(list(data)):
        base_info = DemoInfo.from_space_id(space_id)
        info = DemoInfo(**(dataclasses.asdict(base_info) | data[space_id]))
        demo_info.append(info)
    return pd.DataFrame([dataclasses.asdict(info) for info in demo_info])


class Prettifier:
    @staticmethod
    def create_link(text: str, url: str) -> str:
        return f'<a href={url} target="_blank">{text}</a>'

    @staticmethod
    def to_div(text: str | None, category_name: str) -> str:
        if text is None:
            text = ""
        class_name = f"{category_name}-{text.lower()}"
        return f'<div class="{class_name}">{text}</div>'

    @staticmethod
    def add_div_tag_to_replicas(replicas: int) -> str:
        if replicas == 0:
            return ""
        if replicas == 1:
            return "1"
        return f'<div class="multiple-replicas">{replicas}</div>'

    @staticmethod
    def add_div_tag_to_sleep_time(sleep_time_s: str, hardware: str) -> str:
        if hardware == "cpu-basic":
            return f'<div class="sleep-time-cpu-basic">{sleep_time_s}</div>'
        s = sleep_time_s.replace(" ", "-")
        return f'<div class="sleep-time-{s}">{sleep_time_s}</div>'

    def __call__(self, df: pd.DataFrame) -> pd.DataFrame:
        new_rows = []
        for _, row in df.iterrows():
            new_row = dict(row) | {
                "status": self.to_div(row.status, "status"),
                "hardware": self.to_div(row.hardware, "hardware"),
                "suggested_hardware": self.to_div(row.suggested_hardware, "hardware"),
                "title": self.create_link(row.title, row.url),
                "owner": self.create_link(row.owner, f"https://huggingface.co/{row.owner}"),
                "sdk": self.to_div(row.sdk, "sdk"),
                "sleep_time": (
                    self.add_div_tag_to_sleep_time(SLEEP_TIME_INT_TO_STR[row.sleep_time], row.hardware)
                    if ~np.isnan(row.sleep_time)
                    else ""
                ),
                "replicas": self.add_div_tag_to_replicas(row.replicas),
            }
            new_rows.append(new_row)
        return pd.DataFrame(new_rows, columns=df.columns)


class DemoList:
    COLUMN_INFO = (
        ["featured_week", "str"],
        ["status", "markdown"],
        ["hardware", "markdown"],
        ["title", "markdown"],
        ["owner", "markdown"],
        ["likes", "number"],
        ["last_modified", "str"],
        ["created", "str"],
        ["sdk", "markdown"],
        ["sdk_version", "str"],
        ["suggested_hardware", "markdown"],
        ["sleep_time", "markdown"],
        ["replicas", "markdown"],
    )

    def __init__(self, df: pd.DataFrame) -> None:
        self.df_raw = df
        self._prettifier = Prettifier()
        self.df_prettified = self._prettifier(df).loc[:, self.column_names]

    @property
    def column_names(self) -> list[str]:
        return list(map(operator.itemgetter(0), self.COLUMN_INFO))

    def get_column_datatypes(self, column_names: list[str]) -> list[str]:
        mapping = dict(self.COLUMN_INFO)
        return [mapping[name] for name in column_names]

    def filter(
        self,
        status: list[str],
        hardware: list[str],
        sdk: list[str],
        sleep_time: list[str],
        multiple_replicas: bool,
        owner: str,
        start_date: datetime.datetime,
        end_date: datetime.datetime,
        column_names: list[str],
    ) -> pd.DataFrame:
        df = self.df_raw.copy()

        if multiple_replicas:
            df = df[self.df_raw.replicas > 1]
        if owner != "(ALL)":
            df = df[self.df_raw.owner == owner]

        sleep_time_int = [SLEEP_TIME_STR_TO_INT[s] for s in sleep_time]
        df = df[
            (self.df_raw.status.isin(status))
            & (self.df_raw.hardware.isin(hardware))
            & (self.df_raw.sleep_time.isin(sleep_time_int))
            & (self.df_raw.sdk.isin(sdk))
            & (self.df_raw.featured_week >= start_date)
            & (self.df_raw.featured_week <= end_date)
        ]
        df["featured_week"] = df["featured_week"].dt.strftime("%Y-%m-%d")

        return self._prettifier(df).loc[:, column_names]