Merge pull request #165 from nasa/feature/calc_error_extension

Calc Error Extension

src/progpy/utils/calc_error.py

@@ -139,8 +139,10 @@ def MSE(m, times: List[float], inputs: List[dict], outputs: List[dict], **kwargs
             stability_tol represents the fraction of the provided argument `times` that are required to be met in simulation, 
             before the model goes unstable in order to produce a valid estimate of error.
 
-            If the model goes unstable before stability_tol is met, NaN is returned. 
+            If the model goes unstable before stability_tol is met and short_sim_penalty is  None, then exception is raised
+            Else if the model goes unstable before stability_tol is met and short_sim_penalty is not None- the penalty is added to the score
             Else, model goes unstable after stability_tol is met, the error calculated from data up to the instability is returned.
+        short_sim_penalty (float, optional): penalty added for simulation becoming unstable before stability_tol, added for each % below tol. Default is 100
 
     Returns:
         float: Total error
@@ -180,8 +182,15 @@ def MSE(m, times: List[float], inputs: List[dict], outputs: List[dict], **kwargs
             # This is true for any window-based model
             if any(np.isnan(z_obs.matrix)):
                 if t <= cutoffThreshold:
-                    raise ValueError(f"Model unstable- NAN reached in simulation (t={t}) before cutoff threshold. "
-                                     f"Cutoff threshold is {cutoffThreshold}, or roughly {stability_tol * 100}% of the data")     
+                    short_sim_penalty = kwargs.get('short_sim_penalty', 100)
+                    if short_sim_penalty is None:
+                        raise ValueError(f"Model unstable- NAN reached in simulation (t={t}) before cutoff threshold. "
+                                     f"Cutoff threshold is {cutoffThreshold}, or roughly {stability_tol * 100}% of the data")
+
+                    warn(f"Model unstable- NAN reached in simulation (t={t}) before cutoff threshold. "
+                                     f"Cutoff threshold is {cutoffThreshold}, or roughly {stability_tol * 100}% of the data. Penalty added to score.")
+                    # Return value with Penalty added
+                    return err_total/counter + (100-(t/cutoffThreshold)*100)*short_sim_penalty
                 else:
                     warn("Model unstable- NaN reached in simulation (t={})".format(t))
                     break