Fix printing log function for numeric solver expressions

[pnbabu]

The predefined functions ln() and log(), when used in the equations block in the ODEs, result in a compilation error when a numeric solver is selected.

Currently, the ln() function is passed as log() to the ode-toolbox, which results in the error. According to the sympy docs, log has an alias ln(). Thus, replacing log with ln in the ode-toolbox printer fixes the error.

[github-actions]

Bencher Report

Branch	1250/merge
Testbed	ubuntu-latest

🚨 1 Alert

Iteration	Benchmark	Measure Units	View	Benchmark Result (Result Δ%)	Upper Boundary (Limit %)
0	tests/nest_continuous_benchmarking/test_nest_continuous_benchmarking.py::TestNESTContinuousBenchmarking::test_stdp_nn_synapse	Latency seconds (s)	📈 plot 🚷 threshold 🚨 alert (🔔)	3.84 s (+12.86%) Baseline: 3.41 s	3.75 s (102.60%)

Click to view all benchmark results

Benchmark	Latency	Benchmark Result seconds (s) (Result Δ%)	Upper Boundary seconds (s) (Limit %)
tests/nest_continuous_benchmarking/test_nest_continuous_benchmarking.py::TestNESTContinuousBenchmarking::test_stdp_nn_synapse	📈 view plot 🚷 view threshold 🚨 view alert (🔔)	3.84 s (+12.86%) Baseline: 3.41 s	3.75 s (102.60%)

Benchmark	Latency	Benchmark Result seconds (s) (Result Δ%)	Upper Boundary seconds (s) (Limit %)
tests/nest_continuous_benchmarking/test_nest_continuous_benchmarking.py::TestNESTContinuousBenchmarking::test_stdp_nn_synapse	📈 view plot 🚷 view threshold 🚨 view alert (🔔)	3.24 s (-4.60%) Baseline: 3.40 s	3.74 s (86.73%)

Benchmark	Latency	Benchmark Result seconds (s) (Result Δ%)	Upper Boundary seconds (s) (Limit %)
tests/nest_continuous_benchmarking/test_nest_continuous_benchmarking.py::TestNESTContinuousBenchmarking::test_stdp_nn_synapse	📈 view plot 🚷 view threshold 🚨 view alert (🔔)	3.21 s (-5.38%) Baseline: 3.40 s	3.74 s (86.01%)

🐰 View full continuous benchmarking report in Bencher

[clinssen]

Either log10 or ln should be used to avoid ambiguities about the base (see https://nestml.readthedocs.io/en/latest/nestml_language/nestml_language_concepts.html#predefined-functions). Does using log in a NESTML model (without a custom NESTML model function log being defined) not result in an error?

[pnbabu]

Either log10 or ln should be used to avoid ambiguities about the base (see https://nestml.readthedocs.io/en/latest/nestml_language/nestml_language_concepts.html#predefined-functions). Does using log in a NESTML model (without a custom NESTML model function log being defined) not result in an error?

The model has an ln() function, which gets translated to log while sending to ode-toolbox. The ode-toolbox returns the function as log as well. I checked that we do not have any models till now that use an ln() or log10() function inside the equations block. It is either used in the internals or the update block, or inside a user-defined function.

When the solver returned by the ode-toolbox is analytic, we do not see this error because the printers print the function as log, which is compiled by the C++ compiler. Whereas in the case of a numeric solver, the log is appended to the node. string, which results in a compilation error. Basically, the PredefinedFunctions check doesn't hold good here, as the function log doesn't match with ln or log10.

[clinssen]

Suggested change

	return 'log10({!s})'
	return 'log({!s, 10})'

sympy.log10() does not exist. Do you know why this was working before?

[pnbabu]

This is strange. I also have log10 in the test in this PR, and the test doesn't fail.

[clinssen]

sympy.ln() and sympy.log() are the same function, so this should be equivalent.

[pnbabu]

Yes, I replaced it because of the error when using log functions in ODEs with a numeric solver. See here.