# Machine Learning **URL:** https://discourse.julialang.org/c/domain/ml/24.md?page=27 [Latest](https://discourse.julialang.org/latest.md) · [Categories](https://discourse.julialang.org/categories.md) · [Tags](https://discourse.julialang.org/tags.md) **Page:** 28 --- ## [Backpropagating through orthogonal projection operator](https://discourse.julialang.org/t/backpropagating-through-orthogonal-projection-operator/63083)

**Author:** [@rkube](https://discourse.julialang.org/u/rkube)\ **Replies:** 2\ **Last updated:** [January 25, 2022, 5:32pm UTC](https://discourse.julialang.org/t/backpropagating-through-orthogonal-projection-operator/63083 "2022-01-25T17:32:34Z")

Hi, I have a neural network whose loss functions requires to calculate orthogonal projections onto the output of the network. The code runs fine on the CPU, albeit slow. Porting it to the GPU is difficult and I’m stuck.… --- ## [Missing line from deffeqflux blog post](https://discourse.julialang.org/t/missing-line-from-deffeqflux-blog-post/75141)

**Author:** [@pl-ai](https://discourse.julialang.org/u/pl-ai)\ **Replies:** 1\ **Last updated:** [January 25, 2022, 8:10am UTC](https://discourse.julialang.org/t/missing-line-from-deffeqflux-blog-post/75141 "2022-01-25T08:10:13Z")

In the blog post linked from the diffeqflux package (DiffEqFlux.jl – A Julia Library for Neural Differential Equations), there is this section down the page: Notice that we are not learning a solution to the ODE. Instea… --- ## [Training Deep Neural Network using Data Parallel?](https://discourse.julialang.org/t/training-deep-neural-network-using-data-parallel/75079)

**Author:** [@rkube](https://discourse.julialang.org/u/rkube)\ **Replies:** 5\ **Last updated:** [January 24, 2022, 9:08pm UTC](https://discourse.julialang.org/t/training-deep-neural-network-using-data-parallel/75079 "2022-01-24T21:08:19Z")

Hi, I need to train some simpler networks on large datasets and would like to use multiple GPUs for training. Is there a tutorial or writeup on how to do this in Julia? I know about this one in torch: Multi-GPU Examples… --- ## [Zero gradients with Zygote vs correct gradients with ReverseDiff using DiffEqFlux](https://discourse.julialang.org/t/zero-gradients-with-zygote-vs-correct-gradients-with-reversediff-using-diffeqflux/74398)

**Author:** [@JordiBolibar](https://discourse.julialang.org/u/JordiBolibar)\ **Replies:** 4\ **Last updated:** [January 24, 2022, 11:05am UTC](https://discourse.julialang.org/t/zero-gradients-with-zygote-vs-correct-gradients-with-reversediff-using-diffeqflux/74398 "2022-01-24T11:05:07Z")

I’m having some issues (again) with Zygote trying to differentiate a UDE using DiffEqFlux. I already encountered an error with Zygote gradients with the sqrt function. For this new issue I still haven’t been able to isol… --- ## [Vector of Flux Model Parameters Sharing Underlying Data](https://discourse.julialang.org/t/vector-of-flux-model-parameters-sharing-underlying-data/75091)

**Author:** [@RS-Coop](https://discourse.julialang.org/u/RS-Coop)\ **Replies:** 2\ **Last updated:** [January 24, 2022, 3:12am UTC](https://discourse.julialang.org/t/vector-of-flux-model-parameters-sharing-underlying-data/75091 "2022-01-24T03:12:04Z")

I am trying to extract the parameters of a Flux model in a single vector such that they share the same underlying data, i.e. making changes to the vector changes the model parameters. The following is based on the Flux.d… --- ## [MLJ KNNRegressor BruteTree broken for predict-- load error brutetree has no field indices...?](https://discourse.julialang.org/t/mlj-knnregressor-brutetree-broken-for-predict-load-error-brutetree-has-no-field-indices/74837)

**Author:** [@diadora77](https://discourse.julialang.org/u/diadora77)\ **Replies:** 4\ **Last updated:** [January 22, 2022, 5:02pm UTC](https://discourse.julialang.org/t/mlj-knnregressor-brutetree-broken-for-predict-load-error-brutetree-has-no-field-indices/74837 "2022-01-22T17:02:51Z")

Using MLJ’s NearestNeighbors KNNRegressor and attempted to use BruteTree algorithm. Get the following error when attempting to use predict: ERROR: LoadError: type BruteTree has no field indices followed by usual stack t… --- ## [NEAT implementation using FLux](https://discourse.julialang.org/t/neat-implementation-using-flux/74931)

**Author:** [@SanAlphaTau](https://discourse.julialang.org/u/SanAlphaTau)\ **Replies:** 2\ **Last updated:** [January 21, 2022, 10:00am UTC](https://discourse.julialang.org/t/neat-implementation-using-flux/74931 "2022-01-21T10:00:10Z")

Heya, I am trying to implement a simplified version of the NEAT algorithm in julia for a project. I was just wondering what the best way of mapping my genomes into actual code it. Can I build flux models “neuron by neur… --- ## [Convolutional GRU/LSTM in Flux?](https://discourse.julialang.org/t/convolutional-gru-lstm-in-flux/74770)

**Author:** [@johnbb](https://discourse.julialang.org/u/johnbb)\ **Replies:** 4\ **Last updated:** [January 20, 2022, 8:46pm UTC](https://discourse.julialang.org/t/convolutional-gru-lstm-in-flux/74770 "2022-01-20T20:46:21Z")

Is there any implementation of (2D) convolutional GRU, LSTM or similar based on Flux? A similar question was raised in Sept. 2020. --- ## [Need some help with FluxOpt](https://discourse.julialang.org/t/need-some-help-with-fluxopt/74904)

**Author:** [@stochastic\_guy](https://discourse.julialang.org/u/stochastic_guy)\ **Replies:** 1\ **Last updated:** [January 20, 2022, 6:59pm UTC](https://discourse.julialang.org/t/need-some-help-with-fluxopt/74904 "2022-01-20T18:59:36Z")

I am trying to use the FluxOptTools (https://github.com/baggepinnen/FluxOptTools.jl) to train a Flux model using Optim . I followed the example provided in the readme. It is working as intended. But when I slightly chang… --- ## [Zygote terribly slow. What am I doing wrong?](https://discourse.julialang.org/t/zygote-terribly-slow-what-am-i-doing-wrong/74877)

**Author:** [@shiroghost](https://discourse.julialang.org/u/shiroghost)\ **Replies:** 12\ **Last updated:** [January 19, 2022, 10:39pm UTC](https://discourse.julialang.org/t/zygote-terribly-slow-what-am-i-doing-wrong/74877 "2022-01-19T22:39:18Z")

Hi all, I have been using ReverseDiff.jl for a project for quite some time. Now I have switched to Zygote because I need to determine derivative w.r.t. custom types and this seems straightforward with Zygote (see Evalua… --- ## [Evaluate gradient (backpropagation) on a user defined type](https://discourse.julialang.org/t/evaluate-gradient-backpropagation-on-a-user-defined-type/74434)

**Author:** [@shiroghost](https://discourse.julialang.org/u/shiroghost)\ **Replies:** 4\ **Last updated:** [January 19, 2022, 6:20pm UTC](https://discourse.julialang.org/t/evaluate-gradient-backpropagation-on-a-user-defined-type/74434 "2022-01-19T18:20:02Z")

Hi all, I have the following problem. Given a function f(x::Vector) I know that I can evaluate the gradient using any AD package (In particular I use ReverseDiff.jl). What I need now is to evaluate the gradient on a cus… --- ## [CatViews with Flux optimize!](https://discourse.julialang.org/t/catviews-with-flux-optimize/74759)

**Author:** [@lineycroc](https://discourse.julialang.org/u/lineycroc)\ **Replies:** 4\ **Last updated:** [January 19, 2022, 3:35pm UTC](https://discourse.julialang.org/t/catviews-with-flux-optimize/74759 "2022-01-19T15:35:08Z")

Flux’s update!(opt, x, x̄) function errors when x and x̄ are of type CatView with the following message: TypeError: in typeassert, expected Tuple{CatView{1, Float64}, CatView{1, Float64}, Vector{Float64}}, got a value o… --- ## [How to efficiently build AD-compatible matrices line by line](https://discourse.julialang.org/t/how-to-efficiently-build-ad-compatible-matrices-line-by-line/74632)

**Author:** [@GlenHenshaw](https://discourse.julialang.org/u/GlenHenshaw)\ **Replies:** 17\ **Last updated:** [January 15, 2022, 3:46am UTC](https://discourse.julialang.org/t/how-to-efficiently-build-ad-compatible-matrices-line-by-line/74632 "2022-01-15T03:46:01Z")

I find myself frequently needing a construct like the following: a = Array{Float32}(undef, 100, 5) for i=1:100 a\[i, :\] = \ end which works fine in plain Julia, and is highly efficient, but gives Flux fits … --- ## [Flailing on evaluating batches as matrices of features](https://discourse.julialang.org/t/flailing-on-evaluating-batches-as-matrices-of-features/74446)

**Author:** [@GlenHenshaw](https://discourse.julialang.org/u/GlenHenshaw)\ **Replies:** 6\ **Last updated:** [January 14, 2022, 4:13pm UTC](https://discourse.julialang.org/t/flailing-on-evaluating-batches-as-matrices-of-features/74446 "2022-01-14T16:13:48Z")

This is my first significant Flux program, be gentle :wink: I have a model m. It takes two vectors as input (via a Parallel(vcat...) layer), the first of which is a vector of floats and the second is a one-hot. I can ge… --- ## [Import TensorFlow lite model](https://discourse.julialang.org/t/import-tensorflow-lite-model/74572)

**Author:** [@tpoisot](https://discourse.julialang.org/u/tpoisot)\ **Replies:** 0\ **Last updated:** [January 13, 2022, 7:09pm UTC](https://discourse.julialang.org/t/import-tensorflow-lite-model/74572 "2022-01-13T19:09:23Z")

I have a model that’s available as a .tflite file, and I’d like to retrain it in Julia (preferably using Flux). Is there a standard way of importing a tflite file in Julia? --- ## [LSTM - Scaling training data MinMaxScaler Flux alternative](https://discourse.julialang.org/t/lstm-scaling-training-data-minmaxscaler-flux-alternative/74323)

**Author:** [@Jan\_Svanda](https://discourse.julialang.org/u/Jan_Svanda)\ **Replies:** 2\ **Last updated:** [January 11, 2022, 3:28am UTC](https://discourse.julialang.org/t/lstm-scaling-training-data-minmaxscaler-flux-alternative/74323 "2022-01-11T03:28:37Z")

I try to build a model of a dynamic system very similar to Using PyTorch based LSTM to model nonlinear dynamic systems using Flux in Julia. I need to scale the train/test date to (-1,1) but I didn’t find a scaling functi… --- ## [Help needed with Zygote Flux and custom adjoints on GPU](https://discourse.julialang.org/t/help-needed-with-zygote-flux-and-custom-adjoints-on-gpu/74326)

**Author:** [@TsurHerman](https://discourse.julialang.org/u/TsurHerman)\ **Replies:** 1\ **Last updated:** [January 11, 2022, 3:27am UTC](https://discourse.julialang.org/t/help-needed-with-zygote-flux-and-custom-adjoints-on-gpu/74326 "2022-01-11T03:27:30Z")

I am incorporating some simple non-standard elements in a larger DNN. To achieve this I had to manually define some reverse rules through ChainRulesCore syntax ChainRulesCore.rrule(::typeof(f),args...) Which works gre… --- ## [Implementation of composite quantile loss for Flux](https://discourse.julialang.org/t/implementation-of-composite-quantile-loss-for-flux/73978)

**Author:** [@johnbb](https://discourse.julialang.org/u/johnbb)\ **Replies:** 7\ **Last updated:** [January 9, 2022, 10:55pm UTC](https://discourse.julialang.org/t/implementation-of-composite-quantile-loss-for-flux/73978 "2022-01-09T22:55:27Z")

The composite quantile loss for a set of quantiles can be defined as \\sum\_i \\sum\_ j (p\_j - 1\\{y\_i \< q\_{ji}\\}) (y\_i - q\_{ji}) where p\_j \\in(0,1) denote the j-th quantile level/probability, y\_i the i-th target/observa… --- ## [In MLJ, can I access individual predictions in an \`EnsembleModel\`?](https://discourse.julialang.org/t/in-mlj-can-i-access-individual-predictions-in-an-ensemblemodel/74302)

**Author:** [@ablaom](https://discourse.julialang.org/u/ablaom)\ **Replies:** 1\ **Last updated:** [January 9, 2022, 8:28pm UTC](https://discourse.julialang.org/t/in-mlj-can-i-access-individual-predictions-in-an-ensemblemodel/74302 "2022-01-09T20:28:25Z")