Code

Nn stuff#2

Code

Open

sagarwal-atg wants to merge 178 commits intomis0butter:mainmis0butter/julia_learn:mainfrom CLeARoboticsLab:nn_stuffCLeARoboticsLab/sindy_gp_admm:nn_stuffCopy head branch name to clipboard

+7,010-16,745

Commits

Commits on Apr 30, 2023

• forked from julia_learn. removing everythign that's not GP_june

  mis0butter

  committed
  d9ff1b4

• removed if det(K) == 0 from f_obj in GP_june.jl

  mis0butter

  committed
  2d58516

Commits on May 2, 2023

• updated state plots to accommodate n_vars. working on differentiating eigenvalues ...

  mis0butter

  committed
  af9d01e

Commits on May 5, 2023

• after talk with dfk, updated obj fn to try using trace of kernel, not eigenvalues when det(Ky) = 0

  mis0butter

  committed
  b453020

• starting to create more validation plots

  mis0butter

  committed
  8e62960

Commits on May 26, 2023

• created GaussianSINDy repo, renaming project

  mis0butter

  committed
  8c86cf6

• renamed everything to GaussianSINDy

  mis0butter

  committed
  5e2497e

Commits on Jun 4, 2023

• moved a bunch of functions around. plots are in their own functions, lorenz_test is much simpler than sindy_gp_est6

  mis0butter

  committed
  6e530d7

• init_params works in repo, plot_option is working

  mis0butter

  committed
  f17b4b0

Commits on Jun 5, 2023

• ok made a lot of changes ... moved some fns around, they're in utils now. created a sandbox_test script. removed double parantheses from functions that aren't being optimized. fixing gaussian progr…

  Show description for ce3ccab
  mis0butter

  committed
  ce3ccab

• following documentation for total variational derivative, testing in sandbox

  mis0butter

  committed
  0f5d96c

• committing what I have, creating results

  mis0butter

  committed
  8348b63

Commits on Jun 7, 2023

• thanks very much to Jake for help with evaluating functions of vectors of functions and integrating as ODE, committing

  mis0butter

  committed
  ac33cb0

• OMG able to dynamically create function of vector of functions, as well as integrate

  mis0butter

  committed
  a381cac

• moved build_dx_fn to inside ode_fns.jl

  mis0butter

  committed
  c07af05

• it works with predator_preygit add .

  mis0butter

  committed
  7d847e6

• got the dumb validate script to work for plotting

  mis0butter

  committed
  bc860fe

• ok validate_data works now with ode_sine, had to restart julia repl why

  mis0butter

  committed
  b28b16a

• have results for predator_prey, plots are looking pretty good actually

  mis0butter

  committed
  b10f934

Commits on Jun 8, 2023

• fixed legend bug

  mis0butter

  committed
  8cb6a59

• wtf that was so hard. got latex to show up in plot title

  mis0butter

  committed
  ba77d51

• plot margin looks good now

  mis0butter

  committed
  9e89ec6

• committing state that produced predator_prey plot

  mis0butter

  committed
  bc6df31

• fixed a bad typo. using dx_truth_train instead of dx_fd_train

  mis0butter

  committed
  702ed0f

• created metrics for performance. committing what I have now

  mis0butter

  committed
  c90d37f

Commits on Jun 20, 2023

• put plot of predator_prey into function in utils.jl, starting test/car_hardware_test.jl

  mis0butter

  committed
  f07f666

• created some metrics (just norm) for truth vs sindy vs gpsindy coeffs

  mis0butter

  committed
  fa0ab0c

• building notes in readme. renamed variables to things that make sense in predator_prey_test.jl

  mis0butter

  committed
  440235f

• updated predator prey coeffs and plot fns

  mis0butter

  committed
  56faea1

• merging after pull

  mis0butter

  committed
  4d8045b

Commits on Jun 22, 2023

• updated axis ticks

  mis0butter

  committed
  86de7a3

• omg. I got the margins for plots to look good

  mis0butter

  committed
  015910e

Commits on Jun 23, 2023

• got the legend on the rightgit add src/

  mis0butter

  committed
  e3fa2f2

• switched to RGB colors for plotting

  mis0butter

  committed
  9fbe676

• tried to shove train_test_gpsindy into a function, but takes too long. printing metrics

  mis0butter

  committed
  7496fd5

• committing code that can now generate metrics printout and plots

  mis0butter

  committed
  a40ec17

Commits on Jun 25, 2023

• changed k_fn to k_SE

  mis0butter

  committed
  83d448e

Commits on Jun 28, 2023

• added test/outputs to .gitignore

  mis0butter

  committed
  422a577

• deleted plots (moved out test/outputs)

  mis0butter

  committed
  aa67ab2

Commits on Jun 29, 2023

• modified split_train_test to accommodate cross-validation

  mis0butter

  committed
  bf31056

• plot xtick always includes min and max now

  mis0butter

  committed
  dfa451d

• got training data to plot separately if partitioned

  mis0butter

  committed
  a76ac59

• created plot for just test data

  mis0butter

  committed
  2651bcd

• made plot_predator_prey more efficient

  mis0butter

  committed
  f2ea030

• updating lasso admm plots

  mis0butter

  committed
  487573f

• updated plot_lasso_admm to use latexify

  mis0butter

  committed
  dc557b2

• ugh finally finished admm plot

  mis0butter

  committed
  0e1b977

Commits on Jun 30, 2023

• created k_periodic kernel

  mis0butter

  committed
  c30c29e

• sinusoidal kernel works for posterior test/gaussian_process_test.jl

  mis0butter

  committed
  6459fcd

Commits on Jul 5, 2023

• found reason why validate_data wasn't propagating test plots - pool_data_vecfn needs to have sine terms commented out

  mis0butter

  committed
  376a65a

• gonna not use recursion fns, too confusing to change fn library. using OG SINDy fns

  mis0butter

  committed
  3bf2083

• creating new predator_prey_test2 and SINDy_test to try only truth nonlinear terms in fn library

  mis0butter

  committed
  415ec10

• fixed typo for reducing fn library in pool_data_test

  mis0butter

  committed
  5c89d0c

Commits on Jul 7, 2023

• have batch and plots script running for different dx_noise params

  mis0butter

  committed
  7b5966c

• monte carlo outputs xi_1 and xi_2, creating plots for xi_2

  mis0butter

  committed
  5244701

• added hist to outputs of gpsindy fns, put monte_carlo_gpsindy into GaussianSINDy.jl

  mis0butter

  committed
  0032aa2

Commits on Jul 8, 2023

• added Date, now saving results in folder with date-timestamps

  mis0butter

  committed
  335cfbe

• created vv2m fn and plot for final fn val for each iter

  mis0butter

  committed
  d71b783

• added lambda as input to sindy_gp_admm as tunable parameter

  mis0butter

  committed
  7331fbd

Commits on Jul 9, 2023

• adding rounding to gauss_sample to digits 12, machine precision leaking into kernel making it not symmetric?

  mis0butter

  committed
  6eb4000

• ok, hyperparameter optimization is working for simple Gaussian process case ... so wat going on with ADMM

  mis0butter

  committed
  2972d3b

Commits on Jul 10, 2023

• adding lasso_admm_gp_opt which uses GaussianProcesses.jl to optimize hyperparameters

  mis0butter

  committed
  20c7f9e

• GP hyperparameter optimization is working in sindy_gp_admm_test.jl, but it looks like xi is 0

  mis0butter

  committed
  229746c

• gp hyparameter optimization works really well ...

  mis0butter

  committed
  a14af29

• ok, at this state ... sindy_gp_admm_test is working for j = 1

  mis0butter

  committed
  7595a0b

• added if det(Ky) == 0 in f_obj while loop to regularize

  mis0butter

  committed
  982979a

• committing before updating all fns in GaussianSINDy, oh god ... UNIT TEST UNIT TEST

  mis0butter

  committed
  2355905

• start code cleanup. created opt_fns with opt_xi

  mis0butter

  committed
  3dc781d

• created opt_hp

  mis0butter

  committed
  9c95a2a

• moved shrinkage to opt_fns

  mis0butter

  committed
  bab4c74

• updated opt_xi inputs, simpler hyperparam args

  mis0butter

  committed
  94c19f5

Commits on Jul 11, 2023

• finished admm_lasso, now testing

  mis0butter

  committed
  a637da6

• added relaxation parameter to admm_lasso and rho, idk what rho does

  mis0butter

  committed
  6835674

• simplified hyperparameters as one vector of 3 elements as arg input, implemented LU_inv as matrix inverse

  mis0butter

  committed
  b63fd87

• improved comments and var names

  mis0butter

  committed
  94fd084

• created push_hist fn

  mis0butter

  committed
  2d80fd3

• created gpsindy fn

  mis0butter

  committed
  541a50d

• gpsindy works with loopgit add . yay

  mis0butter

  committed
  14309cb

• created monte carlo loop for gpsindy

  mis0butter

  committed
  56f5cb0

• monte_carlo_gpsindy is working

  mis0butter

  committed
  9e9f7cb

• moved gpsindy fns into gpinsdy_fns.jl, but julia is giving method overwritten warnings

  mis0butter

  committed
  fff76b3

• added hist_nvars_vec to monte carlo output

  mis0butter

  committed
  6c4e875

Commits on Jul 13, 2023

• using GP to smooth derivatives for GPSINDy is not helping ... gonna try just using GP to smooth derivatives and then using SINDy

  mis0butter

  committed
  9138398

Commits on Jul 14, 2023

• removed smoothing derivatives

  mis0butter

  committed
  642e81e

• normalizing data makes it look worse in gpsindy

  mis0butter

  committed
  ce5451a

• deleting old test scripts, not needed

  mis0butter

  committed
  47b8e92

• trying out standardizing data before using GPSINDy. normalizing x and then normalizing dx isn't working great

  mis0butter

  committed
  8abe58b

• put plotting fns inside new script, plot_fns

  mis0butter

  committed
  6e303d2

• moved Xi_true inside stand data case, standardization now looks correct

  mis0butter

  committed
  c895901

Commits on Jul 15, 2023

• omg thank god. GPSINDy is working again. the objval += 1/2 * log( tr(Ky) ) was added back to f_obj

  mis0butter

  committed
  43963a2

• added cases for stand-data-true and stand-data-noisy

  mis0butter

  committed
  2232278

Commits on Jul 17, 2023

• GPSINDy looks better than SINDy for the first time

  mis0butter

  committed
  3f4132f

• deleted old gp test scripts, changed gp_dx --> gp_test and have animations of hp opt

  mis0butter

  committed
  6269f09

Commits on Jul 20, 2023

• for kernel SE, put argument order: l, sigma_f

  mis0butter

  committed
  f3d2931

• optimizing HPs in ADMM now looks like it's working, but adding too much noise makes GP bad for optimizing

  mis0butter

  committed
  ffe016a

Commits on Jul 21, 2023

• added animations and plots for fitting xi for sindy and gpsindy

  mis0butter

  committed
  d76f229

Commits on Jul 23, 2023

• created sandbox for testing standardized vs. non-standardized data. idk how to interpret standardized data results

  mis0butter

  committed
  9e197c6

• added another case to gpsindy monte carlo fn, plotting code is a bit messy

  mis0butter

  committed
  63a1619

Commits on Jul 24, 2023

• committing before changing kernel to use t_train input datain f_obj

  mis0butter

  committed
  f5a98ff

• made lots of changes ... added lots of different kernels and changed gp_test to test them

  mis0butter

  committed
  018013f

Commits on Jul 25, 2023

• fixed SINDy_test library so that it finds the correct coefficients for dynamics (using truth data as unit test)

  mis0butter

  committed
  70509df

• FIXED BUG. parameters output is now correct and standardizing data works for scaling coefficients

  mis0butter

  committed
  cc53494

Commits on Jul 26, 2023

• updated GP_tools GP fns to accommodate inputs of different sizes

  mis0butter

  committed
  221497e

• created sandbox for testing Jake's car data. sindy alone looks super bad lmao

  mis0butter

  committed
  cf47235

• giving up on gpsindy for now, trying to leverage SINDy with GP, updated MC cases to reflect

  mis0butter

  committed
  cadfc28

Commits on Jul 27, 2023

• multivariate gps are amazing. smoothing x --> x_GP, then conditioning dx off of x_GP is quite smooth

  mis0butter

  committed
  1622247

• SINDy with GP looks very good compared to SINDy now ...

  mis0butter

  committed
  12e5667

• ok, OG gpsindy is working (with temporal data) again. going to make changes to have non-temporal kernel

  mis0butter

  committed
  855772c

• plot for comparing sindy and gpsindy outputs every case

  mis0butter

  committed
  6416642

• I think I updated ADMM to use x as kernel input? fixed the super annoying bug

  mis0butter

  committed
  72c18ab

• fixed gpsindy dimension inputs for all cases, no more index 173

  mis0butter

  committed
  5caa403

• removed sanity check println code

  mis0butter

  committed
  2ad4c03

Commits on Jul 28, 2023

• implemented Yue's suggestion of tunning hyperparams in between ADMM iters, it takes forever

  mis0butter

  committed
  c7c1409

• made the quartile plot that took way too long

  mis0butter

  committed
  f49bb36

• comparing with box plots, the continuous box plot plot looks right

  mis0butter

  committed
  a124770

• modified ribbon alpha in median/quartile plot

  mis0butter

  committed
  d37f315

• created test case 9 for gp-sindy-gp-sindy, but 2nd gp mean not new Thetax*xi yet. moved median-quartile plot into plot_fns.jl

  mis0butter

  committed
  af6117e

• added LineSearches to packages so that it avoids the assertion error with HagerZhang

  mis0butter

  committed
  30795ac

Commits on Aug 2, 2023

• comitting this now (in broken state), need to implement hp opt for (y_train - Thetax*xi)

  mis0butter

  committed
  a60e151

• reorganized test folder, added test/old and test/data

  mis0butter

  committed
  52ccc4d

Commits on Aug 7, 2023

• implemented pendulum, julia is dumb with the int and float stuff

  mis0butter

  committed
  7f1ff1e

• moved fn outside of monte_carlo_gpsindy, is now arg input

  mis0butter

  committed
  0f3ce57

Commits on Aug 8, 2023

• added cos to sin terms in fn library - now Xi_true looks right

  mis0butter

  committed
  2f2543e

Commits on Aug 9, 2023

• implemented posterior mean in stand_data_test

  mis0butter

  committed
  ca6652b

• added code from working stand_data_test to gpsindy_fns

  mis0butter

  committed
  55cbd48

• very interesting ... when lowering dt of data, gpsindy and gpsindy_gpsindy > sindy

  mis0butter

  committed
  f380933

• put sindy_gpsindy_gpsindygpsindy into script, can run monte carlos in stand_data_test now

  mis0butter

  committed
  255f9c8

• renamed stand_data_test to gpsindygpsindy_test

  mis0butter

  committed
  b0d8e07

• updated plotting script for gpsindy_gpsindy

  mis0butter

  committed
  3adcc50

• added double_pendulum to ode_fns

  mis0butter

  committed
  b3b5dcf

• created new pendulum_test script, updated pendulum fns to use params inputs

  mis0butter

  committed
  71779d4

• double pendulum animation is working

  mis0butter

  committed
  1ccb51c

Commits on Aug 10, 2023

• finished double pendulum animation

  mis0butter

  committed
  7c7264b

• updated fn name to gpsindy_x2

  mis0butter

  committed
  eaf0cd0

• added test/double-pendulum-chaotic to gitignore

  mis0butter

  committed
  66d18e9

• updated readme for double pendulum chaotic dataset

  mis0butter

  committed
  7908226

• added src/structs.jl

  mis0butter

  committed
  e70b3ec

• updated struct names for Xi to Xi_struct

  mis0butter

  committed
  97f1a42

• gpsindy_x2 now takes xi_hist as input and output

  mis0butter

  committed
  ceba9a9

• added Xi_err_hist

  mis0butter

  committed
  f70b345

• updated gpsindy_x2 to use gp_post with comments

  mis0butter

  committed
  cddd770

Commits on Aug 13, 2023

• ok found issue, standardizing data is making data prop wrong

  mis0butter

  committed
  22ed93f

• fixing merge, stand_data_test stands again

  mis0butter

  committed
  5cae276

• updated readme with research meeting notes

  mis0butter

  committed
  b6c3977

• updated readme formatting

  mis0butter

  committed
  e358ca4

• fixed validation plot problem yay!

  mis0butter

  committed
  8972c4b

Commits on Aug 14, 2023

• added test/data/double-pendulum-chaotic to gitignore

  mis0butter

  committed
  10a78b0

• added chatoic_double_pendulum_csv_test.jl, need to augment function library to accommodate all the weird nonlinear terms. also why are hps going to NaN

  mis0butter

  committed
  db6ff44

• updated gp_post arguments

  mis0butter

  committed
  0719f99

• committing current state, need to work on double pendulum fn library

  mis0butter

  committed
  0bf9fa8

Commits on Aug 15, 2023

• updated monte carlo gpsindy to use gp_post in case 9

  mis0butter

  committed
  089ce8d

• resolved merge conflict

  mis0butter

  committed
  66aea14

Commits on Aug 16, 2023

• there was remaining git conflict

  mis0butter

  committed
  f4e4bcd

• updated readme with Adam notes, emphasized motivating example

  mis0butter

  committed
  309a425

Commits on Aug 17, 2023

• added unicycle dynamics and started car_sim_test

  mis0butter

  committed
  00c1e25

• found issue with unicycle dynamics. need to compute more coupled nonlinear terms in fn library

  mis0butter

  committed
  4c8e7c1

• updated readme with notes. going to update coefficients table for easy interpretability

  mis0butter

  committed
  fb8972b

• added nonlinear combinations of sine functions to fn library

  mis0butter

  committed
  d8c5db1

• updated coefficients string vector, size now matches pool_table

  mis0butter

  committed
  8a6475e

• got pretty tables working for coefficients

  mis0butter

  committed
  6d184d1

• added control inputs to nonlinear terms

  mis0butter

  committed
  9b5468a

Commits on Aug 18, 2023

• updated SINDy_test for optional control input. now can just use one function yay

  mis0butter

  committed
  de30ab3

• using Jake's car data again, understand unicycle dynamics better now

  mis0butter

  committed
  e24d682

Commits on Aug 21, 2023

• reorganized ode_fns into a folder, split across files

  mis0butter

  committed
  bb541b3

Commits on Aug 23, 2023

• updated with note, deadline is ICRA SEPT 15git add .

  mis0butter

  committed
  4b96d6e

• implemented linear and angular acceleration quadcopter dynamics, but confused on control kd stuff. downloaded quadcopter data, but need to convert from long/lat to state

  mis0butter

  committed
  4fe4dd1

Commits on Aug 24, 2023

• updated pretty_coeffs to only include nonzero rows

  mis0butter

  committed
  4262724

• unicycle dynamics finally work with truth, needed to update control input to be same in test script and fn

  mis0butter

  committed
  a4ea89c

• added data_struct, created ode_train_test, reorganized ode fns

  mis0butter

  committed
  32788d1

• updated data_train push in ode_train_test

  mis0butter

  committed
  527f4be

• updated structs for u, working on Jake's hardware data now. need to validate

  mis0butter

  committed
  5e2224c

Commits on Aug 25, 2023

• dealt with data rollovers

  mis0butter

  committed
  5ca1a32

• got the x_GP and dx_GP train/test stuff working for Jake's car data

  mis0butter

  committed
  6e14669

• created plot comparing FD vs GP training data for car

  mis0butter

  committed
  f9f3eb9

Commits on Aug 27, 2023

• that was so confusing ... fixed it build_dx_fn for control input

  mis0butter

  committed
  ef7b103

• CONFIRMED IT WORKS WITH XI_TRUE

  mis0butter

  committed
  d302e1c

Commits on Sep 6, 2023

• pushing latest code. created test/int_control_test.jl for integrating dynamics with control

  mis0butter

  committed
  15ad4ad

Commits on Sep 7, 2023

• Added NN training as a baseline

  

  Shubhankar Agarwal

  committed
  0032d5b