[amss*] Build Warnings Issued by SciPy

[mmcky]

The AMSS lectures are not currently executing through github actions

[mmcky]

Using the latest environment the amss lecture fails with the following error. Perhaps this is an issue with a new version of scipy? (cc: @shizejin)

<ipython-input-8-0cc001bf9680>:18: RuntimeWarning: invalid value encountered in log
  return np.log(c) + self.ψ * np.log(1 - n)
<ipython-input-8-0cc001bf9680>:18: RuntimeWarning: divide by zero encountered in log
  return np.log(c) + self.ψ * np.log(1 - n)
<ipython-input-8-0cc001bf9680>:22: RuntimeWarning: divide by zero encountered in true_divide
  return 1 / c
<ipython-input-4-cc6b33fcda51>:235: RuntimeWarning: invalid value encountered in true_divide
  x * u_c / Eu_c - u_c * (c - T) - Un(c, n) * n - β * xprime,
<ipython-input-4-cc6b33fcda51>:235: RuntimeWarning: invalid value encountered in multiply
  x * u_c / Eu_c - u_c * (c - T) - Un(c, n) * n - β * xprime,
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-9-764875dee216> in <module>
      2 log_example.transfers = True                    # Government can use transfers
      3 log_sequential = SequentialAllocation(log_example)  # Solve sequential problem
----> 4 log_bellman = RecursiveAllocationAMSS(log_example, μ_grid)
      5 
      6 T = 20

<ipython-input-4-cc6b33fcda51> in __init__(self, model, μgrid, tol_diff, tol)
     15 
     16         # Find the first best allocation
---> 17         self.solve_time1_bellman()
     18         self.T.time_0 = True  # Bellman equation now solves time 0 problem
     19 

<ipython-input-4-cc6b33fcda51> in solve_time1_bellman(self)
     62             PF = T(Vf)
     63 
---> 64             Vfnew, policies = self.fit_policy_function(PF)
     65             diff = np.abs((Vf(xgrid) - Vfnew(xgrid)) / Vf(xgrid)).max()
     66 

<ipython-input-4-cc6b33fcda51> in fit_policy_function(self, PF)
     81         cf, nf, xprimef, Tf, Vf = [], [], [], [], []
     82         for s_ in range(S):
---> 83             PFvec = np.vstack([PF(x, s_) for x in self.xgrid]).T
     84             Vf.append(interp(xgrid, PFvec[0, :]))
     85             cf.append(interp(xgrid, PFvec[1:1 + S]))

<ipython-input-4-cc6b33fcda51> in <listcomp>(.0)
     81         cf, nf, xprimef, Tf, Vf = [], [], [], [], []
     82         for s_ in range(S):
---> 83             PFvec = np.vstack([PF(x, s_) for x in self.xgrid]).T
     84             Vf.append(interp(xgrid, PFvec[0, :]))
     85             cf.append(interp(xgrid, PFvec[1:1 + S]))

<ipython-input-4-cc6b33fcda51> in PF(x, s)
    207         '''
    208         if not self.time_0:
--> 209             def PF(x, s): return self.get_policies_time1(x, s, Vf)
    210         else:
    211             def PF(B_, s0): return self.get_policies_time0(B_, s0, Vf)

<ipython-input-4-cc6b33fcda51> in get_policies_time1(self, x, s_, Vf)
    242             bounds = [(0., 100)] * S + [(0., 100)] * S + \
    243                 [self.xbar] * S + [(0., 0.)] * S
--> 244         out, fx, _, imode, smode = fmin_slsqp(objf, self.z0[x, s_],
    245                                               f_eqcons=cons, bounds=bounds,
    246                                               full_output=True, iprint=0,

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/slsqp.py in fmin_slsqp(func, x0, eqcons, f_eqcons, ieqcons, f_ieqcons, bounds, fprime, fprime_eqcons, fprime_ieqcons, args, iter, acc, iprint, disp, full_output, epsilon, callback)
    203                   'args': args}, )
    204 
--> 205     res = _minimize_slsqp(func, x0, args, jac=fprime, bounds=bounds,
    206                           constraints=cons, **opts)
    207     if full_output:

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/slsqp.py in _minimize_slsqp(func, x0, args, jac, bounds, constraints, maxiter, ftol, iprint, disp, eps, callback, finite_diff_rel_step, **unknown_options)
    424 
    425         if mode == -1:  # gradient evaluation required
--> 426             g = append(sf.grad(x), 0.0)
    427             a = _eval_con_normals(x, cons, la, n, m, meq, mieq)
    428 

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py in grad(self, x)
    186         if not np.array_equal(x, self.x):
    187             self._update_x_impl(x)
--> 188         self._update_grad()
    189         return self.g
    190 

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py in _update_grad(self)
    169     def _update_grad(self):
    170         if not self.g_updated:
--> 171             self._update_grad_impl()
    172             self.g_updated = True
    173 

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/_differentiable_functions.py in update_grad()
     89                 self._update_fun()
     90                 self.ngev += 1
---> 91                 self.g = approx_derivative(fun_wrapped, self.x, f0=self.f,
     92                                            **finite_diff_options)
     93 

~/anaconda3/envs/quantecon/lib/python3.8/site-packages/scipy/optimize/_numdiff.py in approx_derivative(fun, x0, method, rel_step, abs_step, f0, bounds, sparsity, as_linear_operator, args, kwargs)
    389 
    390     if np.any((x0 < lb) | (x0 > ub)):
--> 391         raise ValueError("`x0` violates bound constraints.")
    392 
    393     if as_linear_operator:

ValueError: `x0` violates bound constraints.

[shizejin]

Hi @mmcky this looks exact as the issue I helped to solve in amss2 (see QuantEcon/lecture-python-advanced#130). As you said, I think this is caused by the update of scipy in their routine fmin_slsqp. They modified their approximation of Jacobian matrix but it causes numerical nuances that break the code. I am quite confident that we can solve this issue just like last time, by defining a Jacobian approximation function ourselves and passing it to the optimization routine. Should I make a PR to the source code in this repo?

[mmcky]

ah @shizejin you're a genius I wonder if I have migrated the associated py files with the latest changes

[mmcky]

this 17898f1 updates the to use the latest supporting code from lecture-python-advanced

[mmcky]

Thanks @shizejin -- this is now fixed.

[mmcky]

@shizejin these lectures are now running but they do have a number of warnings issued

https://60a31e5802cc6d17a5a6d6f7--wonderful-lalande-528d1c.netlify.app/amss.html

[shizejin]

@mmcky thanks for this message. I am working on refactoring the code and probably I can solve these warnings. Might you let me know if this is in hurry? We plan to wrap up the changes by the end of May.

[mmcky]

@shizejin -- no problem. No hurry. But when you're ready I think the site will be powered by this repo using jupyter-book. Let me know if you need any help with migrating from rst to markdown:myst

[mmcky]

Closing this as warnings may be addressed by upcoming rework of amss