"""Data Cleaner — Hugging Face Space demo. Upload a messy CSV or Excel file, get a cleaned one back. No API key, no account: the file is cleaned inside this Space and only you download the result. """ import tempfile from pathlib import Path import gradio as gr import pandas as pd import spaces def clean_text(value): if isinstance(value, str): return " ".join(value.split()) return value def _is_text(series: pd.Series) -> bool: """Text column? Works for both the old object dtype and pandas 3 str dtype.""" return series.dtype == object or pd.api.types.is_string_dtype(series) def clean_numbers(series: pd.Series) -> pd.Series: """Turn strings like '1 200,50' into numbers, if the column looks numeric.""" if not _is_text(series): return series cleaned = ( series.astype(str) .str.replace("\u00a0", " ", regex=False) .str.replace(" ", "", regex=False) .str.replace(",", ".", regex=False) ) numeric = pd.to_numeric(cleaned, errors="coerce") non_empty = series.notna() & (series.astype(str).str.strip() != "") if non_empty.sum() and numeric[non_empty].notna().mean() >= 0.5: return numeric return series def clean_dates(series: pd.Series) -> pd.Series: """Parse a date-like column to YYYY-MM-DD, if most values parse.""" if not _is_text(series): return series parsed = pd.to_datetime(series, errors="coerce", dayfirst=True, format="mixed") non_empty = series.notna() & (series.astype(str).str.strip() != "") if non_empty.sum() and parsed[non_empty].notna().mean() >= 0.5: return parsed.dt.strftime("%Y-%m-%d") return series def clean(df: pd.DataFrame) -> tuple[pd.DataFrame, dict]: report = {"rows_in": len(df)} df = df.dropna(how="all") report["empty_rows_removed"] = report["rows_in"] - len(df) df = df.apply(lambda col: col.map(clean_text)) for col in df.columns: name = str(col).lower() if "date" in name or "datum" in name: df[col] = clean_dates(df[col]) else: df[col] = clean_numbers(df[col]) before = len(df) df = df.drop_duplicates() report["duplicates_removed"] = before - len(df) report["rows_out"] = len(df) report["empty_cells_left"] = int(df.isna().sum().sum()) return df, report @spaces.GPU def clean_file(upload): if upload is None: return None, "Upload a CSV or Excel file first." path = Path(upload.name) if path.suffix.lower() == ".csv": df = pd.read_csv(path, dtype=str, keep_default_na=False) elif path.suffix.lower() in (".xlsx", ".xls"): df = pd.read_excel(path, dtype=str, keep_default_na=False) else: return None, "Only .csv and .xlsx files are supported." df = df.replace(r"^\s*$", pd.NA, regex=True) cleaned, report = clean(df) out = Path(tempfile.gettempdir()) / f"{path.stem}_clean{path.suffix}" if out.suffix.lower() == ".csv": cleaned.to_csv(out, index=False) else: cleaned.to_excel(out, index=False) text = ( f"Rows: {report['rows_in']} -> {report['rows_out']}\n" f"Empty rows removed: {report['empty_rows_removed']}\n" f"Duplicates removed: {report['duplicates_removed']}\n" f"Empty cells left: {report['empty_cells_left']}" ) return str(out), text with gr.Blocks(title="Data Cleaner") as demo: gr.Markdown( "# Data Cleaner\n" "Upload a messy CSV or Excel file and download the cleaned version. " "Duplicates, empty rows, messy dates and numbers like `1 200,50` " "get fixed in one go. No API key, everything runs here." ) with gr.Row(): file_in = gr.File(label="Upload CSV or Excel", file_types=[".csv", ".xlsx"]) file_out = gr.File(label="Download cleaned file") report = gr.Textbox(label="Report") btn = gr.Button("Clean it") btn.click(clean_file, inputs=file_in, outputs=[file_out, report]) gr.Examples( examples=[["examples/messy.csv"]], inputs=file_in, ) demo.launch()