def interp(df, y_col, id_cols=None):
    if not isinstance(y_col, str):
        raise ValueError("y_col should be string")
    if isinstance(id_cols, str):
        id_cols=[id_cols]
    if id_cols is None:
        id_cols=['__dummyid']
        df=df.with_columns(__dummyid=0)
    lf=df.select(id_cols + [y_col]).lazy()
    value_cols=[x for x in df.columns if x not in id_cols and x!=y_col]
    for value_col in value_cols:
        lf=lf.join(
            df.join_asof(
                df.filter(pl.col(value_col).is_not_null())
                .select(
                    *id_cols, y_col,
                    __value_slope=(pl.col(value_col)-pl.col(value_col).shift().over(id_cols))/(pl.col(y_col)-pl.col(y_col).shift().over(id_cols)), 
                    __value_slope_since=pl.col(y_col).shift(),
                    __value_base=pl.col(value_col).shift()
                    ),
                on=y_col, by=id_cols, strategy='forward'
            )
            .select(
                id_cols+ [y_col] + [pl.coalesce(pl.col(value_col), 
                    pl.coalesce(pl.col('__value_base'), pl.col('__value_base').shift(-1))+
                    pl.coalesce(pl.col('__value_slope'), pl.col('__value_slope').shift(-1))*(pl.col(y_col)-
                    pl.coalesce(pl.col('__value_slope_since'), pl.col('__value_slope_since').shift(-1)))).alias(value_col)]
                )
            .lazy(),
            on=[y_col]+id_cols
            )
    if id_cols[0]=='__dummyid':
        lf=lf.select(pl.exclude('__dummyid'))
    return lf.collect()

interp(df, 'ts')
shape: (4, 2)
┌─────────┬──────────┐
│ ts      ┆ value    │
│ ---     ┆ ---      │
│ f64     ┆ f64      │
╞═════════╪══════════╡
│ 1.0     ┆ 1.0      │
│ 2.01    ┆ 1.63125  │
│ 2.00333 ┆ 1.627081 │
│ 4.2     ┆ 3.0      │
└─────────┴──────────┘