Add MultiObjective Optimization Notebook #155

jonpsy · 2021-05-25T13:08:49Z

No description provided.

review-notebook-app · 2021-05-25T13:08:52Z

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

jonpsy · 2021-05-25T13:12:26Z

Also, since we are waiting for custom inputs from user, I'm not sure how the CI will handle this. Any ideas there?

zoq · 2021-05-25T18:25:31Z

Also, since we are waiting for custom inputs from user, I'm not sure how the CI will handle this. Any ideas there?

I would just go with some predefined parameters and add a comment to let the user know how and what to change.

jonpsy · 2021-05-25T21:22:16Z

@zoq Thanks for the help! Do you have any further queries? I think this notebook is good to go.

zoq · 2021-05-25T21:24:22Z

@zoq Thanks for the help! Do you have any further queries? I think this notebook is good to go.

I'll test it out later, right now I have to manfully pull the latest ensmallen branch right? I have the conda package for ensmallen ready, I just have to do the same thing for mlpack.

zoq · 2021-05-25T21:26:22Z

Maybe it already works, you can add zoq to the channels https://github.com/mlpack/examples/blob/master/binder/environment.yml as well as ensmallen at the end?

jonpsy · 2021-05-25T21:28:59Z

Yeah, I tested it manually and it worked. Anyway the integration test should confirm it for you. I've written a small script to automate the process of pulling the latest ensmallen https://gist.github.com/jonpsy/de2e1f7e49a93a393beadcdae679784a. Just checkout this branch and go to terminal and copy all the code from the script, it will do everything for you. After that, just run this notebook and voila.

Let me know if you find difficulty somewhere?

jonpsy · 2021-05-25T21:35:22Z

Looks like our build broke
The command "./download_data_set.py" failed and exited with 1 during . :)

zoq · 2021-05-25T21:39:15Z

Restarted let's see.

jonpsy · 2021-05-25T21:40:41Z

Maybe it already works, you can add zoq to the channels https://github.com/mlpack/examples/blob/master/binder/environment.yml as well as ensmallen at the end?

channels:
- conda-forge
- zoq
- ensmallen

Something like this?

zoq · 2021-05-25T21:43:23Z

Almost, zoq is correct but ensmallen add the end of the file, since it's the package we like to install.

jonpsy · 2021-05-25T21:46:43Z

Sorry for dragging this on, you mean under dependencies right?

dependencies:
# Jupyter
- jupyterlab=3
- ....
- ensmallen

zoq · 2021-05-25T21:48:16Z

Correct, right below mlpack is good.

jonpsy · 2021-05-26T05:24:46Z

Looks like the test pass 🥳

zoq · 2021-05-26T14:11:28Z

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#define ARMA_DONT_USE_WRAPPER"


It's already part of #include <mlpack/xeus-cling.hpp> so we can remove the extra line here.

zoq · 2021-05-26T14:14:15Z

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "opt.Optimize(objectives, coords);"


Looks like this takes some time, would be a good idea to add some callbacks (https://ensmallen.org/docs.html#callback-documentation) to show some output in between and at the end (PrintLoss, Report).

Looks kinda redundant. PrintLoss() is doing nothing, and Report() has the output as shown in the picture. I think we need to work on MOO callbacks.

Interesting, yes I think we have to adapt the callbacks so that they work for multi-objective optimizers as well.

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

zoq · 2021-05-26T14:19:11Z

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

+   "outputs": [],
+   "source": [
+    "plt::figure_size(800, 800);\n",
+    "plt::scatter(frontX, frontY, 50);\n",


Do you think we could plot the optimization process as well, something like - https://www.machinelearningplus.com/wp-content/uploads/2020/10/ef-300x230.gif

We would need to query the generated Pareto Front per x number of generations, looks like the job of a callback?

Yes, this can be a custom callback, that is implemented as part of the notebook.

BTW, please try and run the notebook once. Our final pareto front isn't as cool as theirs :(.

Agreed, I used https://lab.mlpack.org/ entered your fork and branch to run it, worked pretty well, except that the optimization step took a really long time.

Reg. our results, I do find it kinda odd for the results to be this slow in xeus-cling, my mid-tier laptop can knock it off within seconds. Also, I'm worried about the diversity of the solution, if you recall one of the goals of MOEA is to have a uniform distribution of solutions, what's happening in our algorithm is producing a "cluster" of solutions.

Arguably, since each of the clusters is far apart the solution set is indeed "diverse" so we can't really blame the algorithm. I'm not sure if NSGA-II is to blame or whether our implementation is wrong. It's also possible that our population size is too low to observe the actual Pareto Front but I'm not going to let our users wait an eternity for that.

Yeah, seems odd to me, we should run some more experiments with different parameters. But I think for now this is good, the solution is still valid.

zoq · 2021-05-26T14:20:43Z

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb

+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 5. Final Thoughts"


Maybe it's a good idea to add a section about "How to read the Efficient Frontier?
" see https://www.machinelearningplus.com/machine-learning/portfolio-optimization-python-example/ for an example to the example as well, what do you think?

I've written something similar under the Plotting section. The documentation you suggested explains efficient frontier by comparing it with other fronts, the problem is we can't generate multiple pareto fronts as of now.

I have seen that, but maybe we can improve that section a little bit e.g. by going into some more details what the plot is trying to visualize.

…ipynb Co-authored-by: Marcus Edel <marcus.edel@fu-berlin.de>

This reverts commit af0085e2884af9c5a69af0f8746944d57d5593ba. Added callback

zoq · 2021-05-27T15:09:34Z

@jonpsy Let me know if you like to change anything here, or if this is good to go and we update the notebook in another PR.

jonpsy · 2021-05-27T15:11:06Z

I'll make a few changes, it should be done before kick-off meeting. We can discuss some nitpicks in the meeting and then we can approve by night. Cool?

- Add dominance relation in Optimize section. - Explain X and Y-Axis in Plotting section. - Use the parameters from the blog.

…nto nsga2-notebook

jonpsy · 2021-05-27T19:27:06Z

@zoq Please feel free to merge

zoq

Awesome, thanks for putting this together!

initial push

f910bcb

mlpack-bot bot added s: needs review s: unanswered s: unlabeled labels May 25, 2021

jonpsy commented May 25, 2021

View reviewed changes

portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.ipynb Outdated Show resolved Hide resolved

zoq added c: examples t: added feature and removed s: unanswered s: unlabeled labels May 25, 2021

jonpsy added 2 commits May 26, 2021 02:15

Merge branch 'mlpack:master' into nsga2-notebook

efdabf3

Plotting working! Conclusion done.

da81159

added zoq channel and ensmallen dep

705b131

zoq reviewed May 26, 2021

View reviewed changes

jonpsy and others added 2 commits May 26, 2021 19:55

Update portfolio_optimization_nsga2/portfolio-optimization-nsga2-cpp.…

34b84dd

…ipynb Co-authored-by: Marcus Edel <marcus.edel@fu-berlin.de>

Revert "Added callback"

f18b611

This reverts commit af0085e2884af9c5a69af0f8746944d57d5593ba. Added callback

jonpsy added 2 commits May 27, 2021 21:55

- remove #ARMA_DONT_USE_WRAPPER.

f8bedd4

- Add dominance relation in Optimize section. - Explain X and Y-Axis in Plotting section. - Use the parameters from the blog.

Merge branch 'nsga2-notebook' of https://github.com/jonpsy/examples i…

a6a9971

…nto nsga2-notebook

zoq approved these changes May 27, 2021

View reviewed changes

kartikdutt18 approved these changes May 28, 2021

View reviewed changes

mlpack-bot bot removed the s: needs review label May 28, 2021

kartikdutt18 merged commit a465236 into mlpack:master May 28, 2021

jonpsy deleted the nsga2-notebook branch May 28, 2021 05:16

jonpsy mentioned this pull request May 28, 2021

Framework for Multi-Objective Optimizations - II mlpack/ensmallen#282

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add MultiObjective Optimization Notebook #155

Add MultiObjective Optimization Notebook #155

jonpsy commented May 25, 2021

review-notebook-app bot commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021 •

edited

Loading

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 26, 2021

zoq May 26, 2021

zoq May 26, 2021

jonpsy May 26, 2021

zoq May 26, 2021

zoq May 26, 2021

jonpsy May 26, 2021 •

edited

Loading

zoq May 26, 2021

jonpsy May 26, 2021

zoq May 26, 2021

jonpsy May 27, 2021

zoq May 27, 2021

zoq May 26, 2021

jonpsy May 26, 2021

zoq May 26, 2021

zoq commented May 27, 2021

jonpsy commented May 27, 2021

jonpsy commented May 27, 2021

zoq left a comment

Add MultiObjective Optimization Notebook #155

Add MultiObjective Optimization Notebook #155

Conversation

jonpsy commented May 25, 2021

review-notebook-app bot commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021 • edited Loading

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 25, 2021

zoq commented May 25, 2021

jonpsy commented May 26, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jonpsy May 26, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zoq commented May 27, 2021

jonpsy commented May 27, 2021

jonpsy commented May 27, 2021

zoq left a comment

Choose a reason for hiding this comment

jonpsy commented May 25, 2021 •

edited

Loading

jonpsy May 26, 2021 •

edited

Loading