Better tracking for your deep learning training - Wandb.ai (Weights & Biases)

[@lbiewald]

Hey just wanted to say as one of the founders here at wandb, I really appreciate the video - seeing someone use the product in depth and find it useful makes me very happy. I also appreciate the feedback - we will look into the popup issue and bombard you less with emails.

[@bigdaddy-k2u]

wandb is amazing, thank you for creating it.

[@sathvikudupa1668]

There are lot more features in wandb, it has certainly made my life easier! Do explore more of it!

[@sentdex]

Which ones do you think I should check out?

[@sathvikudupa1668]

@@sentdex reports are neat. you can do all basic doc editing, pull data from your experiments etc. Artifact has dataset versioning. customisations possible in table and Media visualisations are great. They probably have much more to offer, including many small ones like email notifications.

[@tahabouhoun831]

I advocate for reproducibility and being able to create a transparent ML project from data to results. W&B reports are THE embodiment of this #open_science

[@tinylittleanj2]

I love how excited you are about this Harrison… I also didn’t realize I needed this in my life. Thanks for sharing.

[@Iyht]

Hi, I really have to thank you, back in the day when your channel was just starting I was able to find you tutorials here and they really helped me to learn how to code in python. That was like 10 years ago and since then I was able to learn a lot of other programing languages from python to Assembly a C++ and all the in between. I really think that if I wasn't able to find you tutorials back then I would probably had quitted trying to learn to program by myself because back then the learning resources existed but they were very poorly explained and in my opinion they never got to the point to become useful to me. It is the clarity of your explanations and examples that really made programming clicked in my mind and since then whenever I found something new to learn about python or I need to remember how to do something I come back here to re-watch your tutorials. I'm really happy seen the success that your channel had in this years because you really deserve it. Thanks for all the help during this years!!!.

[@adrikovincent5717]

sentdex you are the best, you have taken my learning to another level, thank you

[@DataScienceGarage]

Thanks for sharing it. Using W&B in my projects, really helps to direct the ML developing into the right direction.

[@ChaiTimeDataScience]

Super stoked to see this tutorial!

[@MadlipzMarathi]

sometime I miss your old apartment that drone and framethower looked so cool

[@1337CodeMaster]

Would be really interesting to see you compare multiple other frameworks, like sacred, ClearML or neptune. It's impossible to find a good (non vendor-made) comparison these days.

[@rafail_fridman]

You can also add a Run Comparer table, where you can visualize only the differences in runs' hyperparameters. It is very convenient

[@lvt2050]

It would be very nice to compare all possible products for model tracking, pluses and minuses

[@pennyjohnston8526]

This has been on my todo exploration list. Now using - it's as simple as you say! I have a pipeline comprised of 2 models, this looks to just be for 1 model. I guess I can have a 'linking id' to trace back which combination gave me the best outcome.....or can it do that ?

[@morganmcguire9405]

The parallel coordinates plot (docs.wandb.ai/ref/app/features/panels/parallel-coordinates) would work well for this, where you can add the hyperparameters for each model in each of the vertical axis, e.g. you could visualise the following in the parallel coordinates plot: "model_1_lr, _model_1_dropout, model_2_lr, model_2_wd, final_accuracy" - doing a hyperparameter search over all of these could be visualised nicely If you are training each model separately you could use Artifacts to keep track of which set of model_1 weights was used to train model_2. When saving the model_1 weights to Artifacts you can also log additional metadata along with it. Then when you train model_2 you can pull the weights for model_1 and also log the metadata from model_1 to the wandb run for model 2 - Artifacts (docs.wandb.ai/guides/artifacts) (I work at W&B, hope this makes sense :) )

[@pennyjohnston8526]

@@morganmcguire9405 Thank you so much for this. It sounds just like the functionality I was after - Much appreciated!

[@hassanrevel]

I was thinking about hosting my tensorboard. But they I have got weights and biases, made my work more relaxing.

[@dannychan9461]

14:45 actually I'm stuck at figuring out how to continue training without starting at step 0...

[@anasalrifai2217]

Can you please make a tutorial on how to use its sweep tool for tuning hyper parameters with SB3.

[@starlite5097]

Sentdex: "Really basic hyperparameters" Then there's GPU: "3090" 😳

[@brucemurdock5358]

Just curious. What code font are you using?

[@maciej12345678]

wandb: Network error (ConnectionError), entering retry loop. i have error i nwindows how to fix this

[@sharjeel_mazhar]

Why don't i see a "stop run" option on my end? Whenever i train a model, I don't see any kind of stop run option. Can anyone help me with this?

[@ridaboussioud3005]

hi I have question and I hope you will answer me, I am working on a predictive maintenance project, as an intern in a company, but the dataset that they give me is not labeled ,I mean that there is no feature target that tells us if the machine is failed or not , there is just features like temperature, vibration etc.... my question is can i predict something from this dataset ? if yes how and what I can do ?

[@neuron8186]

i trained on this this is super goodd

[@manikandanshanmugam7480]

Hi, I'm the beginner in ML/AI and Data science. It's really interesting to see your videos. Could you please guide me where do I start?

[@xtra9996]

Hi, since you presented your new ML machine a few weeks ago I've a question: I need a new laptop but I don't want to spend too much money. Now, is a RTX 3070 (mobile) with just 6 GByte of memory sufficient to render ML models or should I buy a standard office laptop without a dedicated accelerator GPU and rent cloud computing CPU / GPU time even for my experiments?

[@iknownotwhoiam]

How does this compare with neptune.ai?

[@SkyFly19853]

Btw, are you going to make new tutorials on Python and Cython? I know that your Python and Cython tutorials are wonderful.

[@dolodestinations7628]

hey there! love the videos. quick question. you have said that a saved model is essentially just weights and biases. and a training model is updating weights and biases by themselves to determine optimization. is there a way to see these weights and biases. so you have 3 columns, is there a way to see the weight of each column for the best performing model?

[@fuba44]

It's not like i check each video making sure my name is in the credits (i totally do that) but this time it looks like I've been left out for some reason, bug in the code? Thinkn I'm gonna try some of the w&b examples, this looks super cool.

[@sentdex]

Hmm, that is indeed odd. I copy and paste the list of members to a text file for the animation to read from, and I confirmed that you were indeed not in the list for that text file. It's possible I just goofed in the copy pasta, but you show up right now in the list of members and I checked the previous animation list and you were on that too. Sorry to have left you out this time!

[@RING_FF]

يارب اتوسل إيليك بحق إسمك الاعضم ان ترزق صاحب اليد التي ستدعــمنـــي فرحه يسجد لها باكيا ورزقا لاينتهي و توفيق النهايه🙌🙌🙌🙌🙌

[@ericlekwane423]

Nueral Networks from scratch ?? Been waiting

[@rafo516]

Bro, I love your laugh

[@ThomasOra-nc7gi]

"Amazon brand" - you're so right 😆

[@emmanotsostrong]

The "ew" from the word "overview" in your thumbnail is obscured by the runtime icon on my subscription page.

[@ultimatums1]

bias about what?

[@justinheehaw]

Oh you're using vscode

[@SPOOKEXE]

noice

[@RING_FF]

يارب اتوسل إيليك بحق إسمك الاعضم ان ترزق صاحب اليد التي ستدعــمنـــي فرحه يسجد لها باكيا ورزقا لاينتهي و توفيق النهايه🙌🙌🙌🙌🙌

[@Stinosko]

Hello

[@peterkovalcik133]

when will the next nnfs video be out if there ever will be one?