nest / ode-toolbox Goto Github PK

See sympy/sympy#23417. This is causing tests to fail for sympy 1.10 and up.

See #19:

Finally, I want to point out that Scipy's odeint does automatic switching between stiff and non-stiff methods and can return method choice and step size information for every step,

sol, info = scipy.integrate.odeint(f, y0, t, (), jac, full_output=True)
step_size_per_step = info['hu']
method_used_per_step = info['mused']
stiff_steps = (method_used_per_step == 2).sum()
non_stiff_steps = sol.size - stiff_steps
# etc

given your API, it would be straightforward to provide this as a fallback when PyGSL is not available (and looking at PyGSL's website does not inspire confidence as a long-term dependency).

This might require a slight redefinition of our solver selection criteria if automatic switching between solvers cannot be disabled in Scipy. Note that Scipy essentially wraps the ODEPACK solver library.

The system of ODEs is

u_bar_plus' = -u_bar_plus / tau
foo' = u_bar_plus / tau + b

The system has a solution when solved with pen and paper, which is as below

u_bar_plus = e^(-t/tau) + c1
foo = -e^(-t/tau) + b.t + c2

However, the output of ode-toolbox is with an undefined value zoo which indicates a bidirectional limit of 1/x with x->0 as mentioned here

"propagators": {
            "__P__foo__foo": "1",
            "__P__foo__u_bar_plus": "1 - exp(-__h/tau)",
            "__P__u_bar_plus__u_bar_plus": "exp(-__h/tau)"
        },
        "solver": "analytical",
        "state_variables": [
            "u_bar_plus",
            "foo"
        ],
        "update_expressions": {
            "foo": "-__P__foo__foo*(b*zoo - foo) - __P__foo__u_bar_plus*(b*zoo - u_bar_plus) + 2*b*zoo",
            "u_bar_plus": "__P__u_bar_plus__u_bar_plus*u_bar_plus"
        }

Between sympy 1.4 and 1.7, the sympy API was changed leading to the error 'sympy.core' has no attribute 'all_classes'.

For now, requirements.txt has been updated to require exactly version 1.4. A more permanent fix should involve rewriting the ODE-toolbox function _is_sympy_type().

Why shape_state_variables and shape_state_updates are stored in the reversed order wrt their derivation order in the json output. But shape_initial_values are sorted by the derivation order? That should be fixed.

To speed up processing, a caching mechanism should be implemented.

See https://ode-toolbox.readthedocs.io/en/latest/index.html#caching-of-results

Due to API change on upstream matplotlib. See https://github.com/timhoffm/matplotlib/blob/9e1eaac2d38749c42ccc51081fad714cd6ac3cdd/doc/api/next_api_changes/removals.rst: "The parameter warn of matplotlib.use has been removed. A failure to switch the backend will now always raise an ImportError if force is set; catch that error if necessary."

Thanks to @shraddhajain13 for reporting this!

ODE-toolbox tries to import matplotlib:

The disadvantage of this is that it (re)sets the plotting backend to Agg. A command-line flag should be added to disable this import.

ode-toolbox globally redirects stderr to /dev/null:

This is a problem when another Python package imports odetoolbox but still wants to use stderr for error reporting (I encountered this bug because antlr4's parse errors during NESTML parsing were not being printed to console).

Consider using sympy common subexpression elimination to make returned expressions less long and repetitive. See: https://docs.sympy.org/latest/modules/rewriting.html

Running the script against the aeif_cond_alpha_implicit.json file in the tests directory raises an error

(NameError: name 'V_m' is not defined).

This must be cached through tests! And test and/or implementation must be fixed.