Updates to regression analysis

    for date, jobs in logdict.items():

        for job, testcases in jobs.items():

            for testcase, value in testcases.items():

                csv_rows.append((job, testcase, date, value))

                formats.append((job, testcase, date, formatdict[date][job][testcase]))

                csv_rows.append((job, testcase, formatdict[date][job][testcase], date, value))

    result = pd.DataFrame(csv_rows, columns=["job", "testcase", "date", "value"])

    result = pd.DataFrame(csv_rows, columns=["job", "testcase", "format", "date", "value"])

    result = result.pivot(

        index=["job", "testcase"], columns="date", values="value"

        index=["job", "testcase", "format"], columns="date", values="value"

    ).reset_index()

    f = pd.DataFrame(formats, columns=["job", "testcase", "date", "format"])

    f = f.pivot(

        index=["job", "testcase"], columns="date", values="format"

    ).reset_index()

    # Keep only tests for which results exist in the last run

    if days == -1:

        rng = result.columns[3:] # Whole history

    else:

        rng = result.columns[-days:]

    result = result.dropna(subset=rng)

    result = result.reset_index(drop=True)

    ratio = result.copy()

    dates = result.iloc[:, 2:].columns

    ratio = ratio.reset_index()

    dates = result.iloc[:, 3:].columns

    # Calculate ratios

    ratio[dates[0]] = 1 # Set first ratio to 1

    ratio[dates[0]] = 1.0 # Set first ratio to 1.0

    for prevdate, currdate in zip( dates[0:-1], dates[1:]):

        ratio[currdate] = result[currdate] / result[prevdate]

    values = result.iloc[:, 2:]

    values = result.iloc[:, 3:]

    date = values.columns

    last_date = date[-1]

    result.insert(2, "format", f[last_date])

    ratio.insert(2, "format", f[last_date])

    formats = result['format'].dropna().unique().tolist()

    plotdata = pd.DataFrame(0.0, index=formats, columns=dates[-(days+1):-1])

    plottext = pd.DataFrame("", index=formats, columns=dates[-(days+1):-1])

    all_indices = []

    for i in range(days):

        currdate = dates[-(days-i+1)] # Make robust for shorter history

        prevdate = dates[-(days-i+2)]

        idx = ratio.groupby("format")[currdate].nlargest(10).index.get_level_values(1)

        tmp = result[["job","testcase","format",prevdate,currdate]].iloc[idx,:].copy().reset_index()

        tmp.insert(3, "prev_date", prevdate)

        tmp.insert(4, "prev_sha", sha[prevdate])

        tmp.insert(5, "curr_date", currdate)

        tmp.insert(6, "curr_sha", sha[prevdate])

        tmp.insert(7, "diff", tmp[currdate] - tmp[prevdate])

        tmp.insert(8, "ratio", tmp[currdate] / tmp[prevdate])

        tmp.loc[tmp[prevdate] == 0, "ratio"] = (

            1  # Set ratio to 1 for denominator 0

        )

        tmp["prev_sha"] = "'" + tmp["prev_sha"] # Add apostrophy to prevent Excel reading this as a number

        tmp["curr_sha"] = "'" + tmp["curr_sha"] # Add apostrophy to prevent Excel reading this as a number

        csv_filename = f"regressions_{measure}_{currdate}.csv"

        tmp.to_csv(csv_filename, sep=";", index=False)

        idx = ratio.groupby("format")[currdate].idxmax()

        all_indices.append(idx)

        # Store worst case per format for plotting

        idx = tmp.groupby("format")["ratio"].idxmax()

        for f in formats:

            plotdata.loc[f, currdate] = tmp.iloc[idx[f]]["ratio"]

            plottext.loc[f, currdate] = f"{tmp.iloc[idx[f]]['job']} - {tmp.iloc[idx[f]]['testcase']} - Max {measure} ratio: {tmp.iloc[idx[f]]['ratio']:.2f}"

            plotdata.loc[f, currdate] = ratio.iloc[idx[f]][currdate]

            plottext.loc[f, currdate] = f"{result.iloc[idx[f]]['job']} - {result.iloc[idx[f]]['testcase']} - Max {measure} ratio: {ratio.iloc[idx[f]][currdate]:.2f}<br>Date: {currdate}"

    fig = make_subplots(

        rows=5,