- Major!! speed improvements
- remote evaluator written in Go
- pulled memoization into search class
- Adds policies for model initialization and expansion
- low: The simplest
- med: The default
- high: This can take much longer
- New Expansion operators
- AddTerm at Toplevel
- Shrink addition operator
- Multi-tiered expansion
- Progressive expansion of models
- Each tier configurable independently
- Adds fitness_function.py
- This allows the fitness calculation for prioritization to be pluggable.
- Motivation for this was to normalize the values across all models.
- Weighted fitness components
- Also added penalized size, jacobian size
- Adds experiments folder, scripts, and config files
- These are related to batch runs and reproducible research.
- Adds timer facilities
- This enables each phase to be timed and can include messages.
- Adds logging facilities
- This enables output to be piped to several, configurable log files.
- Python 3 support
- Missing some, I'm sure...
- Printing models which are a single floating The root cause is the model.subs(...) where an expression simplifies to a single float value. The model.pretty passed to the format string should have been a string. This resulted in an exception that would end the search.
unrecorded...