[bug] No intersection of coverage error and empirical quantile on conformity score

[leoniewgnr]

Describe the bug
When showing the figure of plot_nonconformity_scores in the test test_split_conformal_prediction using the cqr method, the 1-α coverage error (green line) and the empirical quantile q1 (red line), don't meet on the conformity score (blue line).

To Reproduce
Run the following code and add a fig.show() to uncertainty.py to see the plot of fig = plot_nonconformity_scores(self.noncon_scores, self.alpha, self.q_hats[0], method)

def test_split_conformal_prediction():
    log.info("testing: Naive Split Conformal Prediction Air Travel")
    df = pd.read_csv(AIR_FILE)
    m = NeuralProphet(
        seasonality_mode="multiplicative",
        loss_func="MSE",
        quantiles=[0.05, 0.95],
        epochs=EPOCHS,
        batch_size=BATCH_SIZE,
        learning_rate=LR,
    )

    train_df, test_df = m.split_df(df, freq="MS", valid_p=0.2)
    train_df, cal_df = m.split_df(train_df, freq="MS", valid_p=0.15)
    metrics_df = m.fit(train_df, freq="MS")

    alpha = 0.1
    decompose = False
    plotting_backend = "default"
    for method in ["cqr"]: #["naive", "cqr"]:  # Naive and CQR SCP methods
        future = m.make_future_dataframe(
            test_df,
            periods=50,
            n_historic_predictions=len(test_df),
        )
        forecast = m.conformal_predict(
            future,
            calibration_df=cal_df,
            alpha=alpha,
            method=method,
            plotting_backend=plotting_backend,
            decompose=decompose,
        )
        eval_df = uncertainty_evaluate(forecast)

Expected behavior
The intersection of the green and red line, should lie on the blue line.

**Environement **

• Python environment: Python 3.9, in standalone venv with no other packages
• Newest NeuralProphet version cloned from Github
  

[leoniewgnr]

@Kevin-Chen0 did you close that one with #1248?

[leoniewgnr]

closed with #1248