# Machine Learning

**URL:** https://discourse.julialang.org/c/domain/ml/24.md?page=8

[Latest](https://discourse.julialang.org/latest.md) · [Categories](https://discourse.julialang.org/categories.md) · [Tags](https://discourse.julialang.org/tags.md)

**Page:** 9

---

## [Bounds Error when following tutorial of ReinforcementLearning.jl](https://discourse.julialang.org/t/bounds-error-when-following-tutorial-of-reinforcementlearning-jl/113999)

<div class="topic-metadata">

**Author:** [@ras112](https://discourse.julialang.org/u/ras112)\
**Replies:** 0\
**Last updated:** [May 8, 2024, 1:45pm UTC](https://discourse.julialang.org/t/bounds-error-when-following-tutorial-of-reinforcementlearning-jl/113999 "2024-05-08T13:45:17Z")

</div>

Hello community, I have been trying to implement a basic Q-learning algorithm following the tutorial of the ReinforcementLearning.jl. The main issue comes when, instead of giving the agent a RandomPolicy() as policy, I …

---

## [\[English name suggestion\] "FeatureImportanceCalculator", "\[.\]Estimator", "\[.\]Indicator" or what?](https://discourse.julialang.org/t/english-name-suggestion-featureimportancecalculator-estimator-indicator-or-what/113949)

<div class="topic-metadata">

**Author:** [@sylvaticus](https://discourse.julialang.org/u/sylvaticus)\
**Replies:** 3\
**Last updated:** [May 7, 2024, 9:31pm UTC](https://discourse.julialang.org/t/english-name-suggestion-featureimportancecalculator-estimator-indicator-or-what/113949 "2024-05-07T21:31:52Z")

</div>

For a model that returns the ranking of the variables based on some “variable importance metric” to be used as: m = ModelName(a\_ML\_model,options) fit!(m,x,y) var\_importance\_ranking = predict(m) mda\_losses = info(m)\["mda…

---

## [Inconsistent results trying to mesure feature importance](https://discourse.julialang.org/t/inconsistent-results-trying-to-mesure-feature-importance/113885)

<div class="topic-metadata">

**Author:** [@sylvaticus](https://discourse.julialang.org/u/sylvaticus)\
**Replies:** 5\
**Last updated:** [May 6, 2024, 3:23pm UTC](https://discourse.julialang.org/t/inconsistent-results-trying-to-mesure-feature-importance/113885 "2024-05-06T15:23:21Z")

</div>

Hello, I am trying to build a method to generically measure feature importance in ML models but I am having very inconsistent results, that depends on the individual run, and I am very p…off and I don’t understand why. I…

---

## [ReinforcementLearning.jl](https://discourse.julialang.org/t/reinforcementlearning-jl/113745)

<div class="topic-metadata">

**Author:** [@ttidrick](https://discourse.julialang.org/u/ttidrick)\
**Replies:** 0\
**Last updated:** [May 2, 2024, 2:27pm UTC](https://discourse.julialang.org/t/reinforcementlearning-jl/113745 "2024-05-02T14:27:56Z")

</div>

Hi All, I’m considering using RL.jl for a project. I would like to know this discussion group’s opinion about the safety and the usefulness of this tool. Thanks, Ty Tidrick

---

## [Nested AD with Lux etc](https://discourse.julialang.org/t/nested-ad-with-lux-etc/113573)

<div class="topic-metadata">

**Author:** [@avikpal](https://discourse.julialang.org/u/avikpal)\
**Replies:** 26\
**Last updated:** [May 1, 2024, 6:05am UTC](https://discourse.julialang.org/t/nested-ad-with-lux-etc/113573 "2024-05-01T06:05:44Z")

</div>

Nested AD (Starting v0.5.38) Starting v0.5.38, Lux automatically captures some common possibilities of AD calls inside loss functions on lux layers and converts them into a faster version to do a JVP over gradient (inste…

---

## [Flux data parallelism](https://discourse.julialang.org/t/flux-data-parallelism/113627)

<div class="topic-metadata">

**Author:** [@jafagervik](https://discourse.julialang.org/u/jafagervik)\
**Replies:** 2\
**Last updated:** [April 30, 2024, 8:29am UTC](https://discourse.julialang.org/t/flux-data-parallelism/113627 "2024-04-30T08:29:46Z")

</div>

Hi, I’ve tried finding resources regarding Flux/Lux data parallelism. Im still unsure what the current SOTA solution for data parallel training would be. Im creating a different kinds of autoencoders, with the main one …

---

## [Why are Neural SDEs trained with TrackerAdjoint()?](https://discourse.julialang.org/t/why-are-neural-sdes-trained-with-trackeradjoint/113514)

<div class="topic-metadata">

**Author:** [@agelgazzar](https://discourse.julialang.org/u/agelgazzar)\
**Replies:** 1\
**Last updated:** [April 29, 2024, 5:39am UTC](https://discourse.julialang.org/t/why-are-neural-sdes-trained-with-trackeradjoint/113514 "2024-04-29T05:39:31Z")

</div>

I thought that Tracker was mostly replaced by Zygote and Enzyme for Reverse-mode Autodiff in the sciml ecosystem. Why is it still being used for computing the sensitivities in DiffeqFlux?

---

## [Adding external parameters to neural network](https://discourse.julialang.org/t/adding-external-parameters-to-neural-network/112968)

<div class="topic-metadata">

**Author:** [@KianH](https://discourse.julialang.org/u/KianH)\
**Replies:** 5\
**Last updated:** [April 24, 2024, 3:03pm UTC](https://discourse.julialang.org/t/adding-external-parameters-to-neural-network/112968 "2024-04-24T15:03:41Z")

</div>

Is there any way in Lux.jl or Flux.jl to create a function for a UDE and add some parameter, for instance the parameter c in the function rhs!(...) to a neural network so that the parameter c is also estimated through th…

---

## [How to install NeuralOperators as a CPU-Version?](https://discourse.julialang.org/t/how-to-install-neuraloperators-as-a-cpu-version/113283)

<div class="topic-metadata">

**Author:** [@Andrew-Milestone](https://discourse.julialang.org/u/Andrew-Milestone)\
**Replies:** 3\
**Last updated:** [April 22, 2024, 4:52pm UTC](https://discourse.julialang.org/t/how-to-install-neuraloperators-as-a-cpu-version/113283 "2024-04-22T16:52:26Z")

</div>

Hello, i’m new to using julia, but have extensive experience in python and mathematica. unfortunately, i have an old laptop with a non-cuda compatible graphics card. so i’d like to include the NeuralOperators package as…

---

## [Deep learning in Julia](https://discourse.julialang.org/t/deep-learning-in-julia/112844)

<div class="topic-metadata">

**Author:** [@cpfiffer](https://discourse.julialang.org/u/cpfiffer)\
**Replies:** 35\
**Last updated:** [April 22, 2024, 9:11am UTC](https://discourse.julialang.org/t/deep-learning-in-julia/112844 "2024-04-22T09:11:44Z")

</div>

I’m been wondering about the general status of the deep learning ecosystem in Julia. We obviously have the big project Flux.jl, as well as a similar but mutable version in Lux.jl. I bring this up because it would be FAN…

---

## [Energy Time Series GRU with Historic and Forecasted Variables](https://discourse.julialang.org/t/energy-time-series-gru-with-historic-and-forecasted-variables/112986)

<div class="topic-metadata">

**Author:** [@dmoored4](https://discourse.julialang.org/u/dmoored4)\
**Replies:** 1\
**Last updated:** [April 21, 2024, 5:50pm UTC](https://discourse.julialang.org/t/energy-time-series-gru-with-historic-and-forecasted-variables/112986 "2024-04-21T17:50:44Z")

</div>

Hello, all! I need conceptual assistance with how I should be using a GRU which uses historic data for variables 1:8 and forecasted variables for variables 1:4 to forecast the values for variables 5:8. I have time serie…

---

## [One dimensional autoencoder as a way to refine (or evaluate) cluster analysis](https://discourse.julialang.org/t/one-dimensional-autoencoder-as-a-way-to-refine-or-evaluate-cluster-analysis/112875)

<div class="topic-metadata">

**Author:** [@sylvaticus](https://discourse.julialang.org/u/sylvaticus)\
**Replies:** 1\
**Last updated:** [April 20, 2024, 11:54am UTC](https://discourse.julialang.org/t/one-dimensional-autoencoder-as-a-way-to-refine-or-evaluate-cluster-analysis/112875 "2024-04-20T11:54:06Z")

</div>

Is it correct (and do you have references I can use in order to "sit on the giant’s shoulders…) to use 1-D autoencoders output to analyse the heterogeneity of my data, a bit like the cluster analysis but clusters maps th…

---

## [Many breakthroughs: Complex-valued transformer neural networks, or even "quaternion backpropagation", or none at all? Predictive coding](https://discourse.julialang.org/t/many-breakthroughs-complex-valued-transformer-neural-networks-or-even-quaternion-backpropagation-or-none-at-all-predictive-coding/113135)

<div class="topic-metadata">

**Author:** [@Palli](https://discourse.julialang.org/u/Palli)\
**Replies:** 0\
**Last updated:** [April 18, 2024, 10:14am UTC](https://discourse.julialang.org/t/many-breakthroughs-complex-valued-transformer-neural-networks-or-even-quaternion-backpropagation-or-none-at-all-predictive-coding/113135 "2024-04-18T10:14:53Z")

</div>

\[I usually post ML stuff under in “offtopic” but I think I may get better answers/discussion here, I’m not asking for a solution to a very specific problem I have. I start with interesting practical developments then mor…

---

## [GeometricMachineLearning.jl: Looking for a direct contact opportunity with persons involved](https://discourse.julialang.org/t/geometricmachinelearning-jl-looking-for-a-direct-contact-opportunity-with-persons-involved/113025)

<div class="topic-metadata">

**Author:** [@j\_u](https://discourse.julialang.org/u/j_u)\
**Replies:** 0\
**Last updated:** [April 16, 2024, 3:42pm UTC](https://discourse.julialang.org/t/geometricmachinelearning-jl-looking-for-a-direct-contact-opportunity-with-persons-involved/113025 "2024-04-16T15:42:48Z")

</div>

Hi, I am looking for a direct contact opportunity with persons involved in GeometricMachineLearning.jl. I joined Julia Community a few years ago thanks to a hobby project of mine about "Neural network analysis of uncerta…

---

## [GPU memory usage increasing on each epoch (Flux)](https://discourse.julialang.org/t/gpu-memory-usage-increasing-on-each-epoch-flux/112942)

<div class="topic-metadata">

**Author:** [@evanfields](https://discourse.julialang.org/u/evanfields)\
**Replies:** 5\
**Last updated:** [April 16, 2024, 2:16pm UTC](https://discourse.julialang.org/t/gpu-memory-usage-increasing-on-each-epoch-flux/112942 "2024-04-16T14:16:01Z")

</div>

I’m catching up on the last few years of Flux progress (thanks, Flux devs!) by training some toy models, and am surprised to find that GPU memory usage is increasing on each training epoch, even when all training data is…

---

## [NeuralODE with covariates](https://discourse.julialang.org/t/neuralode-with-covariates/112999)

<div class="topic-metadata">

**Author:** [@elcup](https://discourse.julialang.org/u/elcup)\
**Replies:** 1\
**Last updated:** [April 16, 2024, 9:34am UTC](https://discourse.julialang.org/t/neuralode-with-covariates/112999 "2024-04-16T09:34:16Z")

</div>

Hi. Am an quite new with Julia and I am stuck with something. Maybe you can help me (sorry if I am trying something stupid). I am working with Neural ODEs using DiffEqFlux. I want to model the differential equation with…

---

## [ODEProblem(....) vs NeuralODE(....) for neural ODEs](https://discourse.julialang.org/t/odeproblem-vs-neuralode-for-neural-odes/112670)

<div class="topic-metadata">

**Author:** [@KianH](https://discourse.julialang.org/u/KianH)\
**Replies:** 14\
**Last updated:** [April 15, 2024, 2:16pm UTC](https://discourse.julialang.org/t/odeproblem-vs-neuralode-for-neural-odes/112670 "2024-04-15T14:16:22Z")

</div>

Hello! I am wondering about the difference when using ODEProblem(....) from DifferentialEquations.jl and NeuralODE(....) from DiffEqFlux.jl in terms of time performance. Consider the code below, thus, definining the …

---

## [Can I define a type-unstable EnzymeRule?](https://discourse.julialang.org/t/can-i-define-a-type-unstable-enzymerule/112732)

<div class="topic-metadata">

**Author:** [@gdalle](https://discourse.julialang.org/u/gdalle)\
**Replies:** 1\
**Last updated:** [April 9, 2024, 6:09pm UTC](https://discourse.julialang.org/t/can-i-define-a-type-unstable-enzymerule/112732 "2024-04-09T18:09:34Z")

</div>

I am trying to define custom Enzyme rules for my package ImplicitDifferentiation.jl, but I’m held back by type inference failures inside my rules. The typical error message looks like Enzyme execution failed. Enzyme: …

---

## [Lipschitz continuity -- Neural ODEs](https://discourse.julialang.org/t/lipschitz-continuity-neural-odes/108463)

<div class="topic-metadata">

**Author:** [@RahulManavalan](https://discourse.julialang.org/u/RahulManavalan)\
**Replies:** 2\
**Last updated:** [April 7, 2024, 12:13pm UTC](https://discourse.julialang.org/t/lipschitz-continuity-neural-odes/108463 "2024-04-07T12:13:28Z")

</div>

Here’s a open ended topic that I haven’t seen people mention, explicitly in literature. In order to integrate neural ODEs, there needs to be guarentees regarding the existence of its solutions. And any arbitrary NN appr…

---

## [What are \`autodiff\_deferred\` and \`autodiff\_thunk\` for in Enzyme?](https://discourse.julialang.org/t/what-are-autodiff-deferred-and-autodiff-thunk-for-in-enzyme/112065)

<div class="topic-metadata">

**Author:** [@gdalle](https://discourse.julialang.org/u/gdalle)\
**Replies:** 23\
**Last updated:** [April 7, 2024, 5:37am UTC](https://discourse.julialang.org/t/what-are-autodiff-deferred-and-autodiff-thunk-for-in-enzyme/112065 "2024-04-07T05:37:00Z")

</div>

I think I understand the basic autodiff function (see this post) but it has three variants: autodiff\_deferred autodiff\_thunk (which seems broken on 1.11) autodiff\_deferred\_thunk Is there someone who can explain to me: …

---

## [DeepAR in Julia](https://discourse.julialang.org/t/deepar-in-julia/71707)

<div class="topic-metadata">

**Author:** [@wormholestds9](https://discourse.julialang.org/u/wormholestds9)\
**Replies:** 3\
**Last updated:** [April 6, 2024, 10:53am UTC](https://discourse.julialang.org/t/deepar-in-julia/71707 "2024-04-06T10:53:28Z")

</div>

Does anybody know if there is a Julia package for DeepAR from Amazon? Do I just have to make do by calling its Python package?

---

## [Challenges with Zygote Differentiation in Multivariate Mixture Models as input noise of an implicit machine learning method](https://discourse.julialang.org/t/challenges-with-zygote-differentiation-in-multivariate-mixture-models-as-input-noise-of-an-implicit-machine-learning-method/112560)

<div class="topic-metadata">

**Author:** [@josemanuel22](https://discourse.julialang.org/u/josemanuel22)\
**Replies:** 10\
**Last updated:** [April 5, 2024, 2:23pm UTC](https://discourse.julialang.org/t/challenges-with-zygote-differentiation-in-multivariate-mixture-models-as-input-noise-of-an-implicit-machine-learning-method/112560 "2024-04-05T14:23:25Z")

</div>

The following implicit machine learning method works well when the input noise is, for example, a multidimensional Gaussian function sliced\_invariant\_statistical\_loss\_optimized\_2(nn\_model, loader, hparams) @assert l…

---

## [Pseudo-random freeze during Flux training](https://discourse.julialang.org/t/pseudo-random-freeze-during-flux-training/112357)

<div class="topic-metadata">

**Author:** [@CM1](https://discourse.julialang.org/u/CM1)\
**Replies:** 0\
**Last updated:** [March 31, 2024, 4:54pm UTC](https://discourse.julialang.org/t/pseudo-random-freeze-during-flux-training/112357 "2024-03-31T16:54:34Z")

</div>

I am training the parameters of an epidemiology model via Flux. I am able to obtain gradients for the parameters and for a few epochs, the training works well. However, after some more epochs, the training freezes. Weird…

---

## [Strange situation RF Classification: perfect train, test prediction all one category](https://discourse.julialang.org/t/strange-situation-rf-classification-perfect-train-test-prediction-all-one-category/112222)

<div class="topic-metadata">

**Author:** [@sylvaticus](https://discourse.julialang.org/u/sylvaticus)\
**Replies:** 2\
**Last updated:** [March 28, 2024, 1:45pm UTC](https://discourse.julialang.org/t/strange-situation-rf-classification-perfect-train-test-prediction-all-one-category/112222 "2024-03-28T13:45:49Z")

</div>

I am puzzled by an issue with a boolean classification task using RF on a large dimensional dataset (1680 obs x 110 dim) and moderate imbalance (431 vs 1249). train/test partition is random (0.8,0.2) When I train the R…

---

## [MethodError in Custom Flux model](https://discourse.julialang.org/t/methoderror-in-custom-flux-model/112204)

<div class="topic-metadata">

**Author:** [@CM1](https://discourse.julialang.org/u/CM1)\
**Replies:** 1\
**Last updated:** [March 28, 2024, 2:25am UTC](https://discourse.julialang.org/t/methoderror-in-custom-flux-model/112204 "2024-03-28T02:25:19Z")

</div>

Hey everyone, I have been trying to get my custom model running, which should be able to optimize both for a set of paramters and some initial conditions of an ODE model. The ODE’s dynamics are defined in ode(), the mod…

---

## [Why is Flux.destructure type unstable?](https://discourse.julialang.org/t/why-is-flux-destructure-type-unstable/111963)

<div class="topic-metadata">

**Author:** [@Iris\_Allevi](https://discourse.julialang.org/u/Iris_Allevi)\
**Replies:** 6\
**Last updated:** [March 25, 2024, 3:51pm UTC](https://discourse.julialang.org/t/why-is-flux-destructure-type-unstable/111963 "2024-03-25T15:51:04Z")

</div>

I was building a simple model and at some point I needed to “unroll” it to get all the parameters in an array. So I tired with Flux.destructure. I got some type instability, so I checked the documentation and I tried wi…

---

## [Is \`cumsum\` differentiable with Zygote when used in a Flux model?](https://discourse.julialang.org/t/is-cumsum-differentiable-with-zygote-when-used-in-a-flux-model/111738)

<div class="topic-metadata">

**Author:** [@MattSainsbury-Dale](https://discourse.julialang.org/u/MattSainsbury-Dale)\
**Replies:** 2\
**Last updated:** [March 18, 2024, 10:24am UTC](https://discourse.julialang.org/t/is-cumsum-differentiable-with-zygote-when-used-in-a-flux-model/111738 "2024-03-18T10:24:16Z")

</div>

I have a call to cumsum in my Flux model that works as intended in the forward pass, but throws an error when I try to differentiate it. A minimum working example: using Flux n = 5 x = \[rand(1, n), rand(1, n)\] # dumm…

---

## [Trying to put together our own MARL environment](https://discourse.julialang.org/t/trying-to-put-together-our-own-marl-environment/111732)

<div class="topic-metadata">

**Author:** [@artisakov](https://discourse.julialang.org/u/artisakov)\
**Replies:** 0\
**Last updated:** [March 17, 2024, 7:47pm UTC](https://discourse.julialang.org/t/trying-to-put-together-our-own-marl-environment/111732 "2024-03-17T19:47:03Z")

</div>

Hi, I’m from a university student RL team and we are going to make a new multi-agent cooperative-adversarial environment for learning optimal strategies for short-term scheduling of incoming bids (surgical procedures in …

---

## [Loss function in NeuralPDE.jl](https://discourse.julialang.org/t/loss-function-in-neuralpde-jl/109702)

<div class="topic-metadata">

**Author:** [@Vivi](https://discourse.julialang.org/u/Vivi)\
**Replies:** 8\
**Last updated:** [March 8, 2024, 11:30am UTC](https://discourse.julialang.org/t/loss-function-in-neuralpde-jl/109702 "2024-03-08T11:30:59Z")

</div>

Hi, I’m trying to train a PINN using NeuralPDE.jl. I was able to start the training and achieved (at least in my opinion) a quiet small loss \< 1e-7. But when I plotted the network solution against the analytic solution …

---

## [Is there something like the DistributionLambda layer from Keras in Flux?](https://discourse.julialang.org/t/is-there-something-like-the-distributionlambda-layer-from-keras-in-flux/111352)

<div class="topic-metadata">

**Author:** [@ohmsweetohm1](https://discourse.julialang.org/u/ohmsweetohm1)\
**Replies:** 0\
**Last updated:** [March 8, 2024, 12:10pm UTC](https://discourse.julialang.org/t/is-there-something-like-the-distributionlambda-layer-from-keras-in-flux/111352 "2024-03-08T12:10:04Z")

</div>

DistributionLambda I guess it could use Distributions.jl.

[Previous page](https://discourse.julialang.org/c/domain/ml/24.md?page=7)

[Next page](https://discourse.julialang.org/c/domain/ml/24.md?page=9)