jax printer doesn't properly convert sympy Min and Max during lambdify

[dardeshna]

jax can't accept tuples as input to amin and amax the same way that numpy does. this results in the following error when lambdifying a sympy expression with min/max:

TypeError: max requires ndarray or scalar arguments, got <class 'tuple'> at position 0.

sympy/sympy/printing/numpy.py

Lines 218 to 222 in a07d5bc

	def _print_Min(self, expr):
	return '{}(({}), axis=0)'.format(self._module_format(self._module + '.amin'), ','.join(self._print(i) for i in expr.args))
	def _print_Max(self, expr):
	return '{}(({}), axis=0)'.format(self._module_format(self._module + '.amax'), ','.join(self._print(i) for i in expr.args))

i think the fix is simply to wrap the tuple in a jax.numpy.asarray() call (similarly to how and/or are carved out for jax)

[virajvekaria]

    def _print_Min(self, expr):
        return '{}(({}), axis=0)'.format(self._module_format(self._module + '.amin'), ','.join('jax.numpy.asarray({})'.format(self._print(i)) for i in expr.args))

    def _print_Max(self, expr):
        return '{}(({}), axis=0)'.format(self._module_format(self._module + '.amax'), ','.join('jax.numpy.asarray({})'.format(self._print(i)) for i in expr.args))

So should this work??

[virajvekaria]

Hey I have made one more change, and I wished to test it. Please can someone help me with that.

[dardeshna]

yes although i might suggest something like this to remove the redundant parentheses

thanks for looking into this!

    def _print_Min(self, expr):
        return '{}({}.asarray([{}]), axis=0)'.format(self._module_format(self._module + '.amin'), self._module_format(self._module), ','.join(self._print(i) for i in expr.args))

    def _print_Max(self, expr):
        return '{}({}.asarray([{}]), axis=0)'.format(self._module_format(self._module + '.amax'), self._module_format(self._module), ','.join(self._print(i) for i in expr.args))

[virajvekaria]

@dardeshna I'm pretty new to open source, what do I do next? Should I issue a Pull request?

[dardeshna]

Honestly me as well — probably open a PR and look through the contributing guidelines for the project to see the workflow used by sympy?

[virajvekaria]

okay i looked into it, I am currently running all the tests, and then probably open a PR, btw I really need to thank you, sir, this is literally my first issue solved in open source

[virajvekaria]

ouch it fails 15 errors

[virajvekaria]

one of the tests is being detected as trojan on my windows computer, please can someone help

[virajvekaria]

Okay, so it's now three errors only, and one is such that Windows Defender detects it as a trojan, should I allow this file to run by putting it on Defender's whitelist?

[oscarbenjamin]

It is hard to answer about the trojan without more information. It sounds like an overzealous virus checker but hard to be sure.

[virajvekaria]

Also, please can tell someone tell what is _np here??

def get_results_with_scipy(objective, constraints, variables):
    if scipy is not None and np is not None:
        from sympy.solvers.inequalities import _np
        nonpos, rep, xx = _np(constraints, [])
        assert not rep  # only testing nonneg variables
        C, _D = linear_eq_to_matrix(objective, *variables)
        A, B = linear_eq_to_matrix(nonpos, *variables)
        assert _D[0] == 0  # scipy only deals with D = 0

This code is from sympy\solvers\tests\test_simplex

[virajvekaria]

It is hard to answer about the trojan without more information. It sounds like an overzealous virus checker but hard to be sure.

So how do I proceed with this, I shouldn't commit if there's an error right? There is an option to allow this code to run by putting it on whitelist but the question is can this code be trusted

For reference this test is sympy\utilities_compilation\tests\test_compilation.py

[asmeurer]

It looks like that _np function was moved in 43d23ed. I guess that code was never executed in the tests CI. It would be better to not use a private function there if possible.

[asmeurer]

While it's usually best to try to get everything working before committing, if you can't figure something out, it's better to just commit what you have an make a pull request with it, so that others can help you figure out how to fix it. It's much easier for people to help you if your code is pushed up so that they can see it and checkout and try it themselves.