ketozhang
commented
2 days ago Not addressing the issue itself, but this has been quite difficult to debug due to lacking error handling in RTModel. Here I have to understand what ModelSelector() is writing to a new directory Models/, so an actual success is only indicated by the existence of files in this directory.
A lot of the code in RTModel Python looks like this
Notable issues are:
- When C++ executable returns non-zero, an error is printed out. If instead, we raise an Exception here, we get the benfit of both notifying and giving control of error handling to the user.
stdout=subprocess.DEVNULLfurther hides any useful information from the C++ executable. Changing this tocapture_output=Trueand passing stdout and stderr to python logging module would allow users more control over its verbosity.
Example of useful logs in ModelSelector not printed out by default for users
``` ********************* **** RTModel **** ********************* Event name: /tmp/pytest-of-root/pytest-6/test_levmar_solution_shape_PS_0 Number of processors: 4 >>> rtm.ModelSelector("PS") - Selecting models for Single-lens-Single-source fits ****************************************** ********** Model selector ********* ****************************************** This program finds the best independent models of a given class - Event: /tmp/pytest-of-root/pytest-6/test_levmar_solution_shape_PS_0 - Reading options in ModelSelector.ini Number of sigmas used to declare overlap between two models: 3.000000 Number of sigmas in chi square distribution for model acceptance: 1.000000 Maximum number of models reported: 10 - Model code: PS Reading data Reading models ! No models for this class OK ```
I'm happy to contribute error handling code.
valboz
In some microlensing light curves, I am getting RTModel silently failing despite all stdout printing success.
Here an example in PSPL:
The
Models/directory is missingThe logs reports everything is success