Regression: performance statistics + plots

regr_stats.py

from sklearn.metrics import r2_score, mean_squared_error
import matplotlib.lines as mlines
import matplotlib.transforms as mtransforms

def stats(value, pred):
    res = [
      {
        'metric': 'R2',
        'value' : r2_score(value, pred)
      }
    ]
    res.append(
      {
        'metric': 'RMSE',
        'value' : sc.sqrt(mean_squared_error(value, pred))
      }
    )
    res.append(
      {
        'metric': 'N',
        'value' : len(value)
      }
    )
    res = pd.DataFrame(res)
    return res

def ba_plot(x, y, title="", x_lab="sell_price lead", y_lab='sell_price difference', s=2):
    fig, ax = plt.subplots()
    ax.scatter(x=x, y=y, s=s)
    ax.axhline(sc.mean(y), xmax=ax.get_xlim()[1], linestyle='--', color='black')
    ax.axhline(sc.mean(y)+sc.std(y), xmax=ax.get_xlim()[1], linestyle='--', color='red')
    ax.axhline(sc.mean(y)-sc.std(y), xmax=ax.get_xlim()[1], linestyle='--', color='red')
    ax.set_title(title)
    ax.set_xlabel(x_lab)
    ax.set_ylabel(y_lab)
    met = [
      {
        'ba_metric': 'mean diff', 'value': sc.mean(y)
      },
      {
        'ba_metric': 'std diff',  'value': sc.std(y)
      }
    ]
    met = pd.DataFrame(met)
    display(met)
    plt.show()

def bland_altman(true_values, pred_values):
    x = true_values
    y = true_values - pred_values
    y_rel = (true_values-pred_values) / true_values
    return pd.DataFrame(
      {
        'x': x,'y_abs': y, 'y_rel': y_rel,
      }
    )
    
def corr_plot(value, pred, x_lab="", y_lab="", title="", s=2):
    fig, ax = plt.subplots()
    ax.scatter(x=value, y=pred, s=s)
    ax.set_title(title)
    ax.set_xlabel(x_lab)
    ax.set_ylabel(y_lab)
    plt.show()
    
def analysis(df, target='sell_price', pred=None):
    metrics = stats(df[target].values, df[pred].values)
    display(metrics)

    corr_plot(df[target].values, df[pred].values, title='correlation plot', y_lab=pred, x_lab=target)
    ba = bland_altman(df[target], df[pred])
    ba_plot(ba['x'], ba['y_abs'], title="absolute values BA", y_lab="abs. {}-{} difference".format(target, pred))
    ba_plot(ba['x'], ba['y_rel'], title="relative values BA", y_lab="rel. {}-{} difference".format(target, pred))