Handle exceptions¶
By default radCAD raises on the first error, ending the simulation. For long Monte Carlo ensembles or large sweeps, you often want the opposite: keep going, collect partial results, and inspect what failed afterwards.
Continue on failure and collect partial results¶
Set raise_exceptions=False on the Engine:
from radcad import Engine
experiment.engine = Engine(raise_exceptions=False)
experiment.run()
results = experiment.results # e.g. [[{...}, {...}], ..., [{...}, {...}]]
exceptions = experiment.exceptions # DataFrame-friendly list of exception metadata
A failed run returns a partial result: the trajectory up to the timestep where it failed. Successful runs and the good portion of failed runs are preserved.
Inspect the exceptions¶
Each entry in experiment.exceptions describes one run, including the exception, its traceback, and the run's metadata:
import pandas as pd
errors = pd.DataFrame(experiment.exceptions)
print(errors[["simulation", "run", "subset", "exception"]])
# Full traceback for the first failure
print(errors.iloc[0]["traceback"])
The metadata fields are: exception, traceback, simulation, run, subset, timesteps, parameters, and initial_state. On a successful run, exception and traceback are None.
Access results from a stored simulation¶
Results and exceptions are stored on the executable, so you can run a simulation and read its output later:
Debugging tip¶
To get a clean, in-process traceback while developing, run on a single process so errors aren't marshalled across process boundaries: