File:Gradient descent.gif

Gradient_descent.gif(360 × 292 pixels, file size: 1.51 MB, MIME type: image/gif, looped, 50 frames)

Summary

Description
English: Gradient descent is a simple method to find the minimum of a function, where at each iteration a small step is made in the direction of the steepest descent. It tends to get stuck in a local minimum, so it is often run with several initial conditions.
Date
Source https://twitter.com/j_bertolotti/status/1121054414066810881
Author Jacopo Bertolotti
Permission
(Reusing this file)
https://twitter.com/j_bertolotti/status/1030470604418428929

Mathematica 11.0 code

f = Evaluate[x^2 + y^2 + Total@Table[RandomReal[{0,2}] E^(-(((x - RandomReal[{-1, 1}])^2 + (y + RandomReal[{-1, 1}])^2)/(2 RandomReal[{0.1, 0.4}]^2))), {10}]];

step = 0.02;
nstep = 100;
coord = {-1, 1};
pos = {coord[[1]], coord[[2]], f /. {x -> coord[[1]], y -> coord[[2]]}};
evo = Reap[Do[
      subst = MapThread[Rule, {{x, y, z}, pos}];
      dfx = (D[{x, y, f}, x] /. subst);
      dfy = (D[{x, y, f}, y] /. subst);
      tmp = {D[f, x] /. subst, D[f, y] /. subst};
      pos = pos - step {tmp[[1]], tmp[[2]], 0};
      pos[[3]] = Evaluate[f /. {x -> pos[[1]], y -> pos[[2]]}];
      Sow[pos];
      , nstep];][[2, 1]];

coord = {-1, -1};
pos = {coord[[1]], coord[[2]], f /. {x -> coord[[1]], y -> coord[[2]]}};
evo1 = Reap[Do[
      subst = MapThread[Rule, {{x, y, z}, pos}];
      dfx = (D[{x, y, f}, x] /. subst);
      dfy = (D[{x, y, f}, y] /. subst);
      tmp = {D[f, x] /. subst, D[f, y] /. subst};
      pos = pos - step {tmp[[1]], tmp[[2]], 0};
      pos[[3]] = Evaluate[f /. {x -> pos[[1]], y -> pos[[2]]}];
      Sow[pos];
      , nstep];][[2, 1]];

coord = {1, 1};
pos = {coord[[1]], coord[[2]], f /. {x -> coord[[1]], y -> coord[[2]]}};
evo2 = Reap[Do[
      subst = MapThread[Rule, {{x, y, z}, pos}];
      dfx = (D[{x, y, f}, x] /. subst);
      dfy = (D[{x, y, f}, y] /. subst);
      tmp = {D[f, x] /. subst, D[f, y] /. subst};
      pos = pos - step {tmp[[1]], tmp[[2]], 0};
      pos[[3]] = Evaluate[f /. {x -> pos[[1]], y -> pos[[2]]}];
      Sow[pos];
      , nstep];][[2, 1]];

p1 = Table[
   Show[
    Plot3D[f, {x, -1.35, 1.35}, {y, -1.35, 1.35}, Boxed -> False, Axes -> False(*,PlotStyle\[Rule]{Opacity[0.3]}*)],
    Graphics3D[{PointSize[0.03], Point[evo[[j]] ], Thick, Line[evo[[1 ;; j]] ], Point[evo1[[j]] ], Line[evo1[[1 ;; j]]] , Point[evo2[[j]] ], Line[evo2[[1 ;; j]]] }]
    ]
   , {j, 1, nstep, 2}];
ListAnimate[p1]

Licensing

I, the copyright holder of this work, hereby publish it under the following license:
Creative Commons CC-Zero This file is made available under the Creative Commons CC0 1.0 Universal Public Domain Dedication.
The person who associated a work with this deed has dedicated the work to the public domain by waiving all of their rights to the work worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law. You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission.

Captions

Gradient descent with 3 different initial conditions.

Items portrayed in this file

depicts

24 April 2019

image/gif

File history

Click on a date/time to view the file as it appeared at that time.

Date/TimeThumbnailDimensionsUserComment
current08:43, 25 April 2019Thumbnail for version as of 08:43, 25 April 2019360 × 292 (1.51 MB)BertoUser created page with UploadWizard
The following pages on the English Wikipedia use this file (pages on other projects are not listed):

Global file usage

The following other wikis use this file:

Metadata