Improve performance #139

ocots · 2023-06-07T17:28:29Z

@PierreMartinon @BaptisteCbl your mission if you accept it is to improve drastically the performance of our solver, thanks to wise profiling and benchmarking.

BaptisteCbl · 2023-06-08T08:52:03Z

For now we have this kind of results:
Results are on classic goddard, with a grid of size 30.
JUMP resolution

Solving...
This is Ipopt version 3.14.13, running with linear solver MUMPS 5.6.0.

Number of nonzeros in equality constraint Jacobian...:      573
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:     1710

Total number of variables............................:      125
                     variables with only lower bounds:       32
                variables with lower and upper bounds:       93
                     variables with only upper bounds:        0
Total number of equality constraints.................:       93
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0  1.0100000e+00 3.96e-01 1.42e-01  -1.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1  1.0099579e+00 3.82e-01 1.55e+01  -1.7 4.61e-01    -  1.04e-02 3.59e-02f  1
   2  1.0123693e+00 9.05e-03 3.60e+03  -1.7 4.80e-01    -  5.61e-02 9.90e-01f  1
   3  1.0089467e+00 6.48e-03 2.52e+03  -1.7 1.30e-01    -  6.51e-01 2.87e-01h  1
   4  1.0012838e+00 2.56e-03 2.85e+02  -1.7 2.80e-01    -  9.15e-01 9.90e-01h  1
   5  1.0055718e+00 6.94e-04 7.38e+04  -1.7 2.99e-01    -  6.99e-01 9.90e-01f  1
   6  1.0072751e+00 2.21e-03 1.25e+06  -1.7 2.56e-01    -  2.70e-01 1.00e+00f  1
   7  1.0072253e+00 1.23e-03 2.25e+05  -1.7 2.87e-01    -  8.86e-01 1.00e+00h  1
   8  1.0067740e+00 2.09e-04 6.58e-01  -1.7 1.00e-01    -  1.00e+00 1.00e+00h  1
   9  1.0067862e+00 1.61e-07 9.53e+02  -3.8 3.16e-03    -  1.00e+00 1.00e+00h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10  1.0087743e+00 5.46e-04 1.44e-01  -3.8 1.36e-01    -  1.00e+00 1.00e+00h  1
  11  1.0086566e+00 4.36e-05 4.27e-04  -3.8 9.79e-02    -  1.00e+00 1.00e+00h  1
  12  1.0102082e+00 2.49e-04 2.57e+03  -5.7 1.12e-01    -  8.27e-01 1.00e+00h  1
  13  1.0119095e+00 4.94e-04 3.83e+02  -5.7 1.51e-01    -  8.51e-01 7.35e-01h  1
  14  1.0122542e+00 2.19e-04 2.17e+01  -5.7 1.81e-01    -  9.43e-01 6.00e-01h  1
  15  1.0124547e+00 1.52e-04 1.27e-03  -5.7 2.00e-01    -  1.00e+00 1.00e+00f  1
  16  1.0124458e+00 5.21e-06 9.77e-05  -5.7 7.70e-02    -  1.00e+00 1.00e+00h  1
  17  1.0124460e+00 1.31e-07 7.94e-07  -5.7 1.13e-02    -  1.00e+00 1.00e+00h  1
  18  1.0125075e+00 1.19e-05 1.17e+01  -8.6 1.00e-01    -  8.33e-01 7.69e-01h  1
  19  1.0125219e+00 1.25e-05 6.03e+00  -8.6 1.67e-01    -  8.53e-01 7.51e-01h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  20  1.0125259e+00 6.81e-06 1.38e+00  -8.6 9.08e-02    -  1.00e+00 8.70e-01h  1
  21  1.0125263e+00 3.46e-07 7.65e-06  -8.6 4.05e-02    -  1.00e+00 1.00e+00f  1
  22  1.0125262e+00 3.43e-08 6.14e-07  -8.6 1.33e-02    -  1.00e+00 1.00e+00h  1
  23  1.0125262e+00 2.18e-09 4.54e-08  -8.6 3.39e-03    -  1.00e+00 1.00e+00h  1
  24  1.0125262e+00 1.87e-11 4.19e-10  -8.6 3.07e-04    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 24

                                   (scaled)                 (unscaled)
Objective...............:  -1.0125262367726526e+00    1.0125262367726526e+00
Dual infeasibility......:   4.1912912891737240e-10    4.1912912891737240e-10
Constraint violation....:   1.8681001190401503e-11    1.8681001190401503e-11
Variable bound violation:   1.4092401593955658e-37    1.4092401593955658e-37
Complementarity.........:   2.5080751303192126e-09    2.5080751303192126e-09
Overall NLP error.......:   2.5080751303192126e-09    2.5080751303192126e-09


Number of objective function evaluations             = 25
Number of objective gradient evaluations             = 25
Number of equality constraint evaluations            = 25
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 25
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 24
Total seconds in IPOPT                               = 0.057

EXIT: Optimal Solution Found.
Local solution found
Objective value = 1.0125262367726526

JUMP -> MathOptNLPModel NLPModelsIpopt resolution

This is Ipopt version 3.14.13, running with linear solver MUMPS 5.6.0.

Number of nonzeros in equality constraint Jacobian...:      573
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:     1710

Total number of variables............................:      125
                     variables with only lower bounds:       32
                variables with lower and upper bounds:       93
                     variables with only upper bounds:        0
Total number of equality constraints.................:       93
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0  1.0100000e+00 3.96e-01 1.42e-01   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1  1.0130396e+00 9.59e-03 1.15e+04  -1.4 4.74e-01    -  1.09e-02 9.90e-01f  1
   2  1.0122507e+00 9.02e-03 1.08e+04  -1.3 5.04e-02   2.0 1.00e+00 5.95e-02h  1
   3  1.0072459e+00 5.60e-03 6.41e+03  -1.7 7.41e-02   1.5 1.00e+00 3.98e-01h  1
   4  1.0067482e+00 5.08e-03 1.22e+04  -1.1 1.30e+00    -  6.31e-01 9.28e-02h  1
   5  1.0025563e+00 1.69e-03 1.73e+03  -1.9 1.63e-01    -  8.09e-01 8.49e-01h  1
   6  1.0027663e+00 8.58e-05 2.26e+02  -7.7 4.12e-02    -  3.64e-01 9.79e-01h  1
   7  1.0027961e+00 9.25e-07 7.40e+01  -3.8 5.79e-03    -  9.10e-01 9.92e-01h  1
   8  1.0028949e+00 5.73e-07 7.99e+02  -5.0 9.87e-03    -  9.53e-01 1.00e+00h  1
   9  1.0071683e+00 7.52e-04 1.02e+03  -4.6 2.65e-01    -  4.03e-01 8.63e-01f  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10  1.0099600e+00 1.07e-03 9.53e+02  -4.6 3.54e-01    -  2.49e-01 5.26e-01h  1
  11  1.0107794e+00 1.17e-03 8.81e+02  -4.2 1.08e+00    -  3.84e-01 4.66e-01h  1
  12  1.0095781e+00 1.33e-03 9.59e+03  -3.8 2.14e+00    -  1.00e+00 2.49e-01h  1
  13  1.0104206e+00 1.31e-04 5.81e-03  -4.5 1.07e-01    -  1.00e+00 1.00e+00h  1
  14  1.0118314e+00 4.06e-04 3.81e+02  -5.0 1.83e-01    -  1.00e+00 8.03e-01h  1
  15  1.0122730e+00 1.40e-04 2.32e-03  -5.2 1.17e-01    -  1.00e+00 1.00e+00h  1
  16  1.0124602e+00 1.01e-04 1.59e-03  -5.8 1.45e-01    -  1.00e+00 1.00e+00h  1
  17  1.0125015e+00 2.16e-05 3.68e-04  -6.2 1.10e-01    -  1.00e+00 1.00e+00h  1
  18  1.0125225e+00 1.79e-05 2.49e-04  -7.0 1.56e-01    -  1.00e+00 1.00e+00h  1
  19  1.0125258e+00 6.06e-06 6.12e-03  -7.8 8.63e-02    -  1.00e+00 9.99e-01h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  20  1.0125263e+00 3.13e-07 6.75e-06  -8.6 3.89e-02    -  1.00e+00 1.00e+00h  1
  21  1.0125263e+00 6.22e-08 1.07e-06  -9.6 1.74e-02    -  1.00e+00 1.00e+00h  1
  22  1.0125263e+00 7.47e-09 1.64e-07 -11.2 6.34e-03    -  1.00e+00 1.00e+00h  1
  23  1.0125263e+00 3.67e-10 8.95e-09 -11.8 1.36e-03    -  1.00e+00 1.00e+00h  1
  24  1.0125263e+00 1.32e-12 1.79e-10 -12.1 8.11e-05    -  1.00e+00 1.00e+00h  1
  25  1.0125263e+00 3.05e-16 1.04e-11 -12.3 1.30e-06    -  1.00e+00 1.00e+00h  1
  26  1.0125263e+00 2.22e-16 1.44e-11 -12.3 6.90e-09    -  1.00e+00 1.00e+00h  1
  27  1.0125263e+00 2.22e-16 1.35e-13 -12.3 4.61e-10    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 27

                                   (scaled)                 (unscaled)
Objective...............:  -1.0125263397980653e+00    1.0125263397980653e+00
Dual infeasibility......:   1.3533594327169007e-13    1.3533594327169007e-13
Constraint violation....:   2.2204460492503131e-16    2.2204460492503131e-16
Variable bound violation:   9.9162263955676622e-09    9.9162263955676622e-09
Complementarity.........:   5.0227126311592196e-13    5.0227126311592196e-13
Overall NLP error.......:   5.0227126311592196e-13    5.0227126311592196e-13


Number of objective function evaluations             = 28
Number of objective gradient evaluations             = 28
Number of equality constraint evaluations            = 28
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 28
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 27
Total seconds in IPOPT                               = 0.068

EXIT: Optimal Solution Found.

CTDirect OCPModel -> ADNLPModel (v0.5) NLPModelsIpopt resolution

This is Ipopt version 3.14.13, running with linear solver MUMPS 5.6.0.

Number of nonzeros in equality constraint Jacobian...:      574
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:     7875

Total number of variables............................:      125
                     variables with only lower bounds:       31
                variables with lower and upper bounds:       62
                     variables with only upper bounds:        0
Total number of equality constraints.................:       94
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0 -1.0100000e+00 9.00e-01 2.00e+00   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1 -1.0030260e+00 7.00e-01 2.48e+02  -0.7 9.00e-01    -  4.85e-03 2.22e-01f  1
   2 -1.0093386e+00 1.84e-01 1.18e+02  -0.7 7.00e-01   2.0 9.97e-01 7.37e-01h  1
   3 -1.0096326e+00 1.07e-01 1.22e+02  -0.4 3.59e-01   1.5 1.00e+00 4.17e-01h  1
   4 -1.0084386e+00 1.07e-03 2.97e+01  -0.6 3.26e-01    -  8.12e-01 9.90e-01h  1
   5 -1.0080173e+00 8.40e-05 1.77e+01  -2.2 5.92e-02    -  9.57e-01 9.90e-01h  1
   6 -1.0080420e+00 1.73e-07 2.79e+00  -7.7 2.19e-03    -  9.85e-01 9.98e-01h  1
   7 -1.0097714e+00 1.37e-04 3.07e+02  -4.7 7.07e-02    -  7.89e-01 1.00e+00f  1
   8 -1.0124338e+00 9.26e-04 6.15e+00  -4.7 2.00e-01    -  1.00e+00 9.84e-01h  1
   9 -1.0123613e+00 1.24e-04 4.87e-03  -5.3 1.13e-01    -  1.00e+00 1.00e+00h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10 -1.0124870e+00 7.79e-05 4.60e+00  -5.9 1.25e-01    -  1.00e+00 9.86e-01h  1
  11 -1.0125120e+00 1.97e-05 5.85e-01  -6.3 1.20e-01    -  1.00e+00 9.93e-01h  1
  12 -1.0125239e+00 1.49e-05 2.24e-01  -7.1 1.52e-01    -  1.00e+00 9.94e-01h  1
  13 -1.0125262e+00 5.03e-06 1.80e-05  -8.1 8.24e-02    -  1.00e+00 1.00e+00h  1
  14 -1.0125262e+00 3.29e-07 8.64e-05  -8.4 3.54e-02    -  1.00e+00 1.00e+00h  1
  15 -1.0125263e+00 5.03e-08 5.30e-07 -10.0 1.66e-02    -  1.00e+00 1.00e+00h  1
  16 -1.0125263e+00 6.43e-09 8.25e-08 -11.0 5.87e-03    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 16

                                   (scaled)                 (unscaled)
Objective...............:  -1.0125263394411863e+00   -1.0125263394411863e+00
Dual infeasibility......:   8.2471633469491931e-08    8.2471633469491931e-08
Constraint violation....:   1.8099726162290608e-09    6.4261801435705479e-09
Variable bound violation:   8.3128719602365564e-09    8.3128719602365564e-09
Complementarity.........:   6.6254607323767334e-10    6.6254607323767334e-10
Overall NLP error.......:   1.8099726162290608e-09    8.2471633469491931e-08


Number of objective function evaluations             = 17
Number of objective gradient evaluations             = 17
Number of equality constraint evaluations            = 17
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 17
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 16
Total seconds in IPOPT                               = 3.009

EXIT: Optimal Solution Found.

CTDirect OCPModel -> ADNLPModel (v0.6) NLPModelsIpopt resolution

This is Ipopt version 3.14.13, running with linear solver MUMPS 5.6.0.

Number of nonzeros in equality constraint Jacobian...:      574
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:      341

Total number of variables............................:      125
                     variables with only lower bounds:       31
                variables with lower and upper bounds:       62
                     variables with only upper bounds:        0
Total number of equality constraints.................:       94
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0 -1.0100000e+00 9.00e-01 2.00e+00   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
   1 -1.0030260e+00 7.00e-01 2.48e+02  -0.7 9.00e-01    -  4.85e-03 2.22e-01f  1
   2 -1.0093386e+00 1.84e-01 1.18e+02  -0.7 7.00e-01   2.0 9.97e-01 7.37e-01h  1
   3 -1.0096326e+00 1.07e-01 1.22e+02  -0.4 3.59e-01   1.5 1.00e+00 4.17e-01h  1
   4 -1.0084386e+00 1.07e-03 2.97e+01  -0.6 3.26e-01    -  8.12e-01 9.90e-01h  1
   5 -1.0080173e+00 8.40e-05 1.77e+01  -2.2 5.92e-02    -  9.57e-01 9.90e-01h  1
   6 -1.0080420e+00 1.73e-07 2.79e+00  -7.7 2.19e-03    -  9.85e-01 9.98e-01h  1
   7 -1.0097714e+00 1.37e-04 3.07e+02  -4.7 7.07e-02    -  7.89e-01 1.00e+00f  1
   8 -1.0124338e+00 9.26e-04 6.15e+00  -4.7 2.00e-01    -  1.00e+00 9.84e-01h  1
   9 -1.0123613e+00 1.24e-04 4.87e-03  -5.3 1.13e-01    -  1.00e+00 1.00e+00h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10 -1.0124870e+00 7.79e-05 4.60e+00  -5.9 1.25e-01    -  1.00e+00 9.86e-01h  1
  11 -1.0125120e+00 1.97e-05 5.85e-01  -6.3 1.20e-01    -  1.00e+00 9.93e-01h  1
  12 -1.0125239e+00 1.49e-05 2.24e-01  -7.1 1.52e-01    -  1.00e+00 9.94e-01h  1
  13 -1.0125262e+00 5.03e-06 1.80e-05  -8.1 8.24e-02    -  1.00e+00 1.00e+00h  1
  14 -1.0125262e+00 3.28e-07 1.04e-06  -8.4 3.53e-02    -  1.00e+00 1.00e+00h  1
  15 -1.0125263e+00 5.01e-08 5.29e-07  -9.9 1.66e-02    -  1.00e+00 1.00e+00h  1
  16 -1.0125263e+00 6.48e-09 8.31e-08 -11.7 5.91e-03    -  1.00e+00 1.00e+00h  1
  17 -1.0125263e+00 2.83e-10 3.33e-09 -12.3 1.20e-03    -  1.00e+00 1.00e+00h  1
  18 -1.0125263e+00 7.57e-13 8.43e-12 -12.3 6.11e-05    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 18

                                   (scaled)                 (unscaled)
Objective...............:  -1.0125263392620425e+00   -1.0125263392620425e+00
Dual infeasibility......:   8.4322266966417110e-12    8.4322266966417110e-12
Constraint violation....:   1.5765166949677223e-13    7.5696393597723964e-13
Variable bound violation:   9.9161631406108341e-09    9.9161631406108341e-09
Complementarity.........:   5.9181867333975404e-13    5.9181867333975404e-13
Overall NLP error.......:   5.9181867333975404e-13    8.4322266966417110e-12


Number of objective function evaluations             = 19
Number of objective gradient evaluations             = 19
Number of equality constraint evaluations            = 19
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 19
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 18
Total seconds in IPOPT                               = 8.586

EXIT: Optimal Solution Found.

To have these results I used the code in : https://github.com/BaptisteCbl/JumpVSADNLP

ocots · 2023-06-12T17:12:39Z

A bench:

using BenchmarkTools

function test_alloc_good()

    function get_view(Y, i, n_slice, N)
        return @view Y[i:i+n_slice-1]
    end

    function set!(Z, i, n_slice, N, z)
        Z[i:i+n_slice-1] = z
    end

    function core_for!(v, x, z, n)
        (v[i] = 2x[i]+3z[i] for i ∈ 1:n)
    end

    n_slice = 50
    n_disc  = 2000
    N = n_slice * n_disc
    Y = rand(N)
    Z = zeros(N)
    x = zeros(n_slice)
    v = zeros(n_slice)

    for i ∈ 1:n_disc

        #
        z = get_view(Y, i, n_slice, N)

        #
        core_for!(v, x, z, n_slice)
        set!(Z, i, n_slice, N, v)

        #
        x[:] = z

    end

    #
    nothing

end

function test_alloc_bad()

    function get(Y, i, n_slice, N)
        return Y[i:i+n_slice-1]
    end

    function set!(Z, i, n_slice, N, z)
        Z[i:i+n_slice-1] = z[:]
    end

    function core(x, z)
        return 2x[:]+3z[:]
    end

    #function core_for(x, z, n)
    #    return [2x[i]+3z[i] for i ∈ 1:n]
    #end

    #function core_for!(v, x, z, n)
    #    v[:] = [2x[i]+3z[i] for i ∈ 1:n]
    #end

    n_slice = 50
    n_disc  = 2000
    N = n_slice * n_disc
    Y = rand(N)
    Z = zeros(N)
    x = zeros(n_slice)
    v = zeros(n_slice)

    for i ∈ 1:n_disc

        #
        z = get(Y, i, n_slice, N)

        #
        set!(Z, i, n_slice, N, core(x[:], z[:]))

        #
        x = z

    end

    #
    nothing

end

println("Allocations and times for good and bad code")
println()

println("good code")
t_good = @benchmark test_alloc_good()
display(t_good)

println()
println("bad code")
t_bad = @benchmark test_alloc_bad()
display(t_bad)

println()
# print the ratio of the times
println("ratio of times: ", mean(t_bad.times)/mean(t_good.times))
println("ratio of allocations: ", mean(t_bad.memory)/mean(t_good.memory))

Results:

Allocations and times for good and bad code

good code
BenchmarkTools.Trial: 3534 samples with 1 evaluation.
 Range (min … max):  332.106 μs … 84.731 ms  ┊ GC (min … max):  0.00% … 73.71%
 Time  (median):       1.050 ms              ┊ GC (median):     0.00%
 Time  (mean ± σ):     1.398 ms ±  2.073 ms  ┊ GC (mean ± σ):  13.85% ± 13.54%

  ▇▂  ▅█▅▄▃▂▁▁     ▁▁                                          ▁
  ███▆████████▇▇█▇█████▆▆▇▅▅▅▆▄▅▅▁▄▅▃▅▄▅▄▇▆▆▅▃▆▇▃▅▃▄▅▅▅▅▅▆▅▅▃▅ █
  332 μs        Histogram: log(frequency) by time      8.57 ms <

 Memory estimate: 1.53 MiB, allocs estimate: 6.

bad code
BenchmarkTools.Trial: 1699 samples with 1 evaluation.
 Range (min … max):  1.623 ms … 81.262 ms  ┊ GC (min … max):  0.00% …  0.00%
 Time  (median):     1.993 ms              ┊ GC (median):     0.00%
 Time  (mean ± σ):   2.916 ms ±  3.547 ms  ┊ GC (mean ± σ):  12.48% ± 17.07%

  █▅▄▃▂▂▃▃▂▃▃▂▂▁ ▁                                            
  ████████████████████▇▇▇▆▆▅▆▇▅▃▅▃▆▃▆▅▅▄▅▄▁▄▄▃▄▁▁▁▁▃▄▁▁▄▁▁▁▅ █
  1.62 ms      Histogram: log(frequency) by time     11.9 ms <

 Memory estimate: 10.04 MiB, allocs estimate: 18006.

ratio of times: 2.0862225595372297
ratio of allocations: 6.5762081784386615

ocots · 2023-06-12T20:15:23Z

@jbcaillau @PierreMartinon @BaptisteCbl @joseph-gergaud

nlp_constraints from CTBase must be updated: https://github.com/control-toolbox/CTBase.jl/blob/bench/bench/bench_nlp_constraints.md

The code: https://github.com/control-toolbox/CTBase.jl/blob/bench/bench/bench_nlp_constraints.jl

ocots · 2024-07-04T20:17:10Z

Too old.

jbcaillau · 2024-07-04T20:34:03Z

@ocots @BaptisteCbl @amontoison well, take a look 🙂

10x speed up wrt to last year version
4x to 5x difference with pure JuMP code

At this price, more than acceptable decrease in performance... and no RK scheme to hard (hand) code.

NB. Tested with ma57 instead of MUMPS.

Method = (:direct, :adnlp, :ipopt)
This is Ipopt version 3.14.14, running with linear solver ma57.

Number of nonzeros in equality constraint Jacobian...:      634
Number of nonzeros in inequality constraint Jacobian.:        0
Number of nonzeros in Lagrangian Hessian.............:      341

Total number of variables............................:      125
                     variables with only lower bounds:       31
                variables with lower and upper bounds:       62
                     variables with only upper bounds:        0
Total number of equality constraints.................:       94
Total number of inequality constraints...............:        0
        inequality constraints with only lower bounds:        0
   inequality constraints with lower and upper bounds:        0
        inequality constraints with only upper bounds:        0

iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
   0 -1.0100000e+00 9.00e-01 2.00e+00   0.0 0.00e+00    -  0.00e+00 0.00e+00   0
Reallocating memory for MA57: lfact (8303)
   1 -1.0030260e+00 7.00e-01 2.48e+02  -0.7 9.00e-01    -  4.85e-03 2.22e-01f  1
   2 -1.0093386e+00 1.84e-01 1.18e+02  -0.7 7.00e-01   2.0 9.97e-01 7.37e-01h  1
   3 -1.0096326e+00 1.07e-01 1.22e+02  -0.4 3.59e-01   1.5 1.00e+00 4.17e-01h  1
   4 -1.0084386e+00 1.07e-03 2.97e+01  -0.6 3.26e-01    -  8.12e-01 9.90e-01h  1
   5 -1.0080173e+00 8.40e-05 1.77e+01  -2.2 5.92e-02    -  9.57e-01 9.90e-01h  1
   6 -1.0080420e+00 1.73e-07 2.79e+00  -7.7 2.19e-03    -  9.85e-01 9.98e-01h  1
   7 -1.0097714e+00 1.37e-04 3.07e+02  -4.7 7.07e-02    -  7.89e-01 1.00e+00f  1
   8 -1.0124338e+00 9.26e-04 6.15e+00  -4.7 2.00e-01    -  1.00e+00 9.84e-01h  1
   9 -1.0123613e+00 1.24e-04 4.87e-03  -5.3 1.13e-01    -  1.00e+00 1.00e+00h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  10 -1.0124870e+00 7.79e-05 4.60e+00  -5.9 1.25e-01    -  1.00e+00 9.86e-01h  1
  11 -1.0125120e+00 1.97e-05 5.85e-01  -6.3 1.20e-01    -  1.00e+00 9.93e-01h  1
  12 -1.0125239e+00 1.49e-05 2.24e-01  -7.1 1.52e-01    -  1.00e+00 9.94e-01h  1
  13 -1.0125262e+00 5.03e-06 1.80e-05  -8.1 8.24e-02    -  1.00e+00 1.00e+00h  1
  14 -1.0125262e+00 3.29e-07 8.64e-05  -8.4 3.54e-02    -  1.00e+00 1.00e+00h  1
  15 -1.0125263e+00 5.03e-08 5.30e-07 -10.0 1.66e-02    -  1.00e+00 1.00e+00h  1
  16 -1.0125263e+00 6.43e-09 8.25e-08 -11.0 5.87e-03    -  1.00e+00 1.00e+00h  1

Number of Iterations....: 16

                                   (scaled)                 (unscaled)
Objective...............:  -1.0125263394411863e+00   -1.0125263394411863e+00
Dual infeasibility......:   8.2471633885825539e-08    8.2471633885825539e-08
Constraint violation....:   1.8099727272513633e-09    6.4261801574483357e-09
Variable bound violation:   8.3128719602365564e-09    8.3128719602365564e-09
Complementarity.........:   6.6254607323533790e-10    6.6254607323533790e-10
Overall NLP error.......:   1.8099727272513633e-09    8.2471633885825539e-08


Number of objective function evaluations             = 17
Number of objective gradient evaluations             = 17
Number of equality constraint evaluations            = 17
Number of inequality constraint evaluations          = 0
Number of equality constraint Jacobian evaluations   = 17
Number of inequality constraint Jacobian evaluations = 0
Number of Lagrangian Hessian evaluations             = 16
Total seconds in IPOPT                               = 0.230

EXIT: Optimal Solution Found.
BenchmarkTools.Trial: 19 samples with 1 evaluation.
 Range (min … max):  233.690 ms … 614.121 ms  ┊ GC (min … max): 0.00% … 55.63%
 Time  (median):     243.806 ms               ┊ GC (median):    1.65%
 Time  (mean ± σ):   267.470 ms ±  84.652 ms  ┊ GC (mean ± σ):  7.51% ± 12.59%

   █                                                             
  ▇█▄▆▆▄▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄ ▁
  234 ms           Histogram: frequency by time          614 ms <

 Memory estimate: 38.34 MiB, allocs estimate: 373614.

PierreMartinon · 2024-07-05T06:34:47Z

Necrothread :D
Still, this is nice. We should add somewhere in the doc the steps to install HSL properly.

jbcaillau · 2024-07-23T13:24:32Z

@ocots @PierreMartinon Can also try to use StaticArrays.jl, most probably in CTFlows.jl as CTDirect.jl are probably too large (size <= 100). Check JuliaArrays/StaticArrays.jl#506

ocots · 2024-07-23T13:34:51Z

Should be used by the user when calling a flow I guess.

jbcaillau · 2024-08-06T13:24:36Z

Related matter: control-toolbox/CTBenchmarks.jl#25

github-actions bot assigned ocots Jun 7, 2023

ocots closed this as completed Jul 4, 2024

jbcaillau reopened this Jul 4, 2024

ocots added the enhancement New feature or request label Jul 26, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve performance #139

Improve performance #139

ocots commented Jun 7, 2023

BaptisteCbl commented Jun 8, 2023 •

edited

Loading

ocots commented Jun 12, 2023

ocots commented Jun 12, 2023 •

edited

Loading

ocots commented Jul 4, 2024

jbcaillau commented Jul 4, 2024 •

edited

Loading

PierreMartinon commented Jul 5, 2024 •

edited

Loading

jbcaillau commented Jul 23, 2024 •

edited

Loading

ocots commented Jul 23, 2024

jbcaillau commented Aug 6, 2024

Improve performance #139

Improve performance #139

Comments

ocots commented Jun 7, 2023

BaptisteCbl commented Jun 8, 2023 • edited Loading

ocots commented Jun 12, 2023

ocots commented Jun 12, 2023 • edited Loading

ocots commented Jul 4, 2024

jbcaillau commented Jul 4, 2024 • edited Loading

PierreMartinon commented Jul 5, 2024 • edited Loading

jbcaillau commented Jul 23, 2024 • edited Loading

ocots commented Jul 23, 2024

jbcaillau commented Aug 6, 2024

BaptisteCbl commented Jun 8, 2023 •

edited

Loading

ocots commented Jun 12, 2023 •

edited

Loading

jbcaillau commented Jul 4, 2024 •

edited

Loading

PierreMartinon commented Jul 5, 2024 •

edited

Loading

jbcaillau commented Jul 23, 2024 •

edited

Loading