Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

quad: Import sympy only when necessary #459

Merged
merged 4 commits into from
Dec 14, 2018
Merged

quad: Import sympy only when necessary #459

merged 4 commits into from
Dec 14, 2018

Conversation

rht
Copy link
Contributor

@rht rht commented Dec 11, 2018 •
edited
Loading

This partially addresses fixes #449. The import quantecon time on my machine decreases from ~5.4s to 4.6s to ~1s.

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

With the second commit, it goes from ~4.6s to ~2s.

@coveralls
Copy link

coveralls commented Dec 11, 2018 •
edited
Loading

Coverage Status

Coverage decreased (-0.01%) to 94.538% when pulling e265b25 on rht:import into ab0c261 on QuantEcon:master.

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

nose-timer stat:

[success] 14.59% quantecon.tests.test_gridtools.test_performance_C: 31.6635s
[success] 10.93% quantecon.tests.test_gridtools.test_performance_F: 23.7086s
[success] 3.17% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_bohachevsky: 6.8868s
[success] 2.98% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_zakharov: 6.4733s
[success] 2.66% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_sum_squared: 5.7616s
[success] 2.50% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_styblinski_tang: 5.4286s
[success] 2.42% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_perm_function: 5.2586s
[success] 2.42% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_easom: 5.2527s
[success] 2.35% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_rotated_hyper_ellipsoid: 5.1085s
[success] 2.28% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_goldstein_price: 4.9372s
[success] 2.26% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_powell: 4.9080s
[success] 2.20% quantecon.util.tests.test_numba.TestNumbaLinalgSolve.test_b_1dim: 4.7764s
[success] 2.18% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_rosenbrock: 4.7321s
[success] 2.16% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_corner_sol: 4.6921s
[success] 2.13% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_colville: 4.6129s
[success] 2.12% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_mccormick: 4.6027s
[success] 2.10% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_booth: 4.5612s
[success] 2.09% quantecon.optimize.tests.test_nelder_mead.TestMaximization.test_discontinuous: 4.5425s
[success] 1.75% quantecon.optimize.tests.test_nelder_mead.test_invalid_σ: 3.7913s
[success] 1.74% quantecon.optimize.tests.test_nelder_mead.test_invalid_ρχ: 3.7754s
[success] 1.71% quantecon.optimize.tests.test_nelder_mead.test_invalid_γ: 3.7168s
[success] 1.63% quantecon.optimize.tests.test_nelder_mead.test_invalid_ρ: 3.5275s
[success] 1.60% quantecon.optimize.tests.test_nelder_mead.test_invalid_χ: 3.4731s
[success] 1.34% quantecon.util.tests.test_numba.TestNumbaLinalgSolve.test_b_2dim: 2.9008s
[success] 1.26% test_lemke_howson.TestLemkeHowson.test_lemke_howson: 2.7230s
[success] 1.20% test_repeated_game.TestAS.test_abreu_sannikov: 2.5982s
[success] 1.15% test_random.TestRandomDiscreteDP.test_equal_beta: 2.5023s
[success] 1.15% quantecon.tests.test_inequality.test_lorenz_curve: 2.4939s
[success] 1.07% test_mclennan_tourky.TestEpsilonNash.test_epsilon_nash_with_full_output: 2.3290s
[success] 0.94% test_random.test_random_stochastic_matrix_dense_vs_sparse: 2.0447s
[success] 0.86% quantecon.tests.test_compute_fp.TestComputeFPContraction.test_num_iter_large: 1.8700s
[success] 0.83% quantecon.optimize.tests.test_nelder_mead.test_invalid_bounds_1: 1.7947s
[success] 0.78% test_support_enumeration.TestSupportEnumeration.test_no_error_skew_sym: 1.6925s
[success] 0.76% quantecon.tests.test_lss.TestLinearStateSpace.test_replicate: 1.6590s
[success] 0.69% test_support_enumeration.TestSupportEnumeration.test_support_enumeration: 1.5076s
[success] 0.67% test_utilities.test_sample_without_replacement_shape: 1.4527s
[success] 0.66% test_utilities.TestProbvec.test_parallel_cpu: 1.4344s
[success] 0.64% quantecon.tests.test_inequality.test_gini_coeff: 1.3894s
[success] 0.58% test_random.test_random_markov_chain_dense: 1.2546s
[success] 0.58% test_normal_form_game.test_player_corner_cases: 1.2525s
[success] 0.53% test_random.TestRandomDiscreteDP.test_shape: 1.1552s
[success] 0.53% test_utilities.TestProbvec.test_shape: 1.1430s
[success] 0.49% test_vertex_enumeration.TestVertexEnumeration.test_vertex_enumeration: 1.0575s
[success] 0.48% test_random.TestRandomDiscreteDP.test_nonzero: 1.0440s
[success] 0.48% quantecon.tests.test_compute_fp.TestComputeFPContraction.test_2d_input: 1.0360s
[success] 0.48% test_random.TestRandomDiscreteDP.test_equal_reward: 1.0338s
[success] 0.48% test_random.TestRandomDiscreteDP.test_equal_probability: 1.0319s
[success] 0.46% quantecon.util.tests.test_timing.TestTicTacToc.test_loop: 1.0033s
[success] 0.44% test_bimatrix_generators.TestTournamentGame.test_payoff_values: 0.9591s
[success] 0.43% test_ddp.test_ddp_beta_0: 0.9365s
[success] 0.36% quantecon.optimize.tests.test_root_finding.test_brentq_basic: 0.7826s
[success] 0.36% quantecon.optimize.tests.test_scalar_max.test_brent_max: 0.7812s
[success] 0.36% test_ddp.TestDiscreteDP.test_modified_policy_iteration: 0.7807s
[success] 0.35% quantecon.optimize.tests.test_root_finding.test_newton_basic: 0.7537s
[success] 0.33% quantecon.optimize.tests.test_scalar_max.test_invalid_a_brent_max: 0.7132s
[success] 0.33% test_normal_form_game.test_best_response_2p: 0.7081s
[success] 0.33% test_random.test_random_stochastic_matrix_sparse: 0.7054s
[success] 0.32% test_random.test_random_stochastic_matrix_dense: 0.6928s
[success] 0.30% test_random.test_random_markov_chain_sparse: 0.6608s
[success] 0.28% quantecon.optimize.tests.test_scalar_max.test_invalid_a_b_brent_max: 0.6117s
[success] 0.28% quantecon.optimize.tests.test_root_finding.test_newton_hard: 0.6071s
[success] 0.27% test_bimatrix_generators.TestBlottoGame.test_constant_diagonal: 0.5965s
[success] 0.26% quantecon.optimize.tests.test_scalar_max.test_invalid_b_brent_max: 0.5594s
[success] 0.22% quantecon.optimize.tests.test_root_finding.test_bisect_basic: 0.4668s
[success] 0.21% quantecon.optimize.tests.test_root_finding.test_halley_hard: 0.4654s
[success] 0.20% quantecon.optimize.tests.test_root_finding.test_halley_basic: 0.4394s

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

test_nelder_mead alone took 46.45% of the entire time. Since Travis can have 4 concurrent jobs, it would be possible to split the test into the cartesian product of [linux, darwin] and [test_nelder_mead, the rest].

@oyamad
Copy link
Member

oyamad commented Dec 11, 2018

Do we really need these tests

[success] 14.59% quantecon.tests.test_gridtools.test_performance_C: 31.6635s
[success] 10.93% quantecon.tests.test_gridtools.test_performance_F: 23.7086s

?

Ref #109

oyamad
oyamad reviewed Dec 11, 2018
View reviewed changes
Copy link
Member

@oyamad oyamad left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change (in the second commit) makes one of the _probvec_* functions complied for every call, which I am afraid is inefficient.

@oyamad
Copy link
Member

oyamad commented Dec 11, 2018 •
edited
Loading

Adding target='parallel' cache=True (CORRECTED) to the _probvec_* functions reduced the import time from 1.922s to 1.003s on the second time import on my machine.

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

OK, I removed the guvectorize commit. I suppose I should have timed the hot start that has the numba caches ready.

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

Adding target='parallel' to the _probvec_* functions reduced the import time from 1.922s to 1.003s on the second time import on my machine.

@oyamad did you mean cache=True?

@rht
Copy link
Contributor Author

rht commented Dec 11, 2018

On quantecon.arma, either:

  1. not put it on the top-level name space (scipy.signal is not at the scipy namespace either, https://github.com/scipy/scipy/blob/master/scipy/__init__.py)
  2. lazy-import the scipy.signal at the class initialization

@oyamad
Copy link
Member

oyamad commented Dec 11, 2018

@rht Yes, cache=True, sorry missed copy-paste...

@oyamad
Copy link
Member

oyamad commented Dec 12, 2018

+1 for 2. lazy import of scipy.signal.

@rht
Copy link
Contributor Author

rht commented Dec 12, 2018

OK, the import time is now at ~1s on my machine,
2018-12-12-110234_1863x900_scrot
Other than the numba import, the rest consists of the cached generated_jit and guvectorize, which I suppose can't be optimized further.

oyamad
oyamad approved these changes Dec 12, 2018
View reviewed changes
Copy link
Member

@oyamad oyamad left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Well done, this is very helpful!

@mmcky
Copy link
Contributor

mmcky commented Dec 13, 2018

thanks @oyamad and @rht this is really great work! That's a pretty neat tool.

@mmcky
Copy link
Contributor

mmcky commented Dec 13, 2018

I think selective imports works well -- when done like this. Is there any PEP guidance to this?

@rht
Copy link
Contributor Author

rht commented Dec 13, 2018

The only article I found documenting various options to do lazy import: https://snarky.ca/lazy-importing-in-python-3-7/ (note: the author wrote importlib.util.LazyLoader for the stdlib, python/cpython@a04dbe4). From the article, I suppose the demand for the functionality was substantial enough to have it formalized in importlib instead of resorting the local imports method I have used here.

@rht
Copy link
Contributor Author

rht commented Dec 13, 2018

See also: numpy/numpy#11457.

@mmcky
Copy link
Contributor

mmcky commented Dec 14, 2018

thanks @rht - this is a great improvement! Also thanks for those links. Lazy importing makes a lot of sense.

@mmcky mmcky merged commit 0a34310 into QuantEcon:master Dec 14, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@oyamad oyamad oyamad approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Startup time is slow
4 participants
@rht @coveralls @oyamad @mmcky