Explorations and Testing Ideas - Brain Computer Interface w/ Python, OpenBCI, and EEG data p.2

Running through where I am currently at, what I've learned, and what I've had trouble with while using the OpenBCI headset for reading EEG data.
openbci.com/
Channel membership: / @sentdex
Discord: / discord
Support the content: pythonprogramming.net/support...
Twitter: / sentdex
Instagram: / sentdex
Facebook: / pythonprogramming.net
Twitch: / sentdex

[@darshanh9197]

This is really helpful... I'm working on project for epileptic seizure predictions through EEG data.. I had no idea about collecting EEG data from headband. Really excited for next video.. 🤩

[@MistaGobo]

I think I'm seeing a rise in the graph data just before/during the moment you're forming a thought - It seems totally correlated with your dialogue. That's cool.

[@N77b44]

As a slightly skeptical take on the graph data spikes, the points where you move your eyebrows seem to sometimes co-occur with when you emphasize a new idea or thought. Muscles work with electrical signals in similar ways to neurons and since they're so much closer to the electrodes the signal from the forehead, blinking, and swallowing generally create large enough electric potentials to make spikes on the graph.

[@jamesjones2212]

@@N77b44 Its still predicting the thought despite it doing it for other reasons

[@shubhanksaxena171]

Imagine moving stuff with this. Finally, we all can be Professor Charles Xavier! Great content man

[@lucasb.2410]

I always enjoy everything that you can make, one day it's finance the other is neuro science, you bring so much fun (&Science) in the game man thank you so much!

[@jaredbeckwith]

Great video, they taught me about Nyquist theorem in EEG college

[@neelgandhi355]

Another outstanding video for use of bci technology

[@freaknoizer1997]

Very enlightener about this bci project !, im about to start a project with this board! So thanks for the video and i hope you can do more vds about this subject!

[@MicaleAntonio]

One of the key pieces of info to using this data well is understanding what each of the frequency ranges represent. Focus on the 15-20 hz range (beta waves) -- this will represent conscious intention better than the other EEG frequencies

[@N77b44]

EEG signals don't tend to do a great job at picking up specific thought contents like left vs right. You might have more success if you focus on modelling a difference between brain patterns in moving two different muscles (e.g. left vs right foot) or in some more global thought difference like hyper focus vs induced boredom. EEG sums over the electrical potentials of neurons across the whole brain so I'd expect the hyper-specific differences in content like "think left" vs "think right" to largely average out considering that those get processed in relatively similar areas for a signal this spatially course. Thinking about or actually moving body areas on the Left vs the Right is "lateralized" in the brain in that the left hemisphere controls the right side and the right hemisphere controls the left side. That's extremely spatially separated so you probably wouldn't get too much signal aggregation and have easily separable conditions.

[@N77b44]

Reward prediction error models might also be pretty cool since the neuro and computer science RL literature really starts to converge around that topic. I think there's a bunch of stuff robustly relating eeg signal to reward prediction error processing in the brain

[@jamesjones2212]

@@N77b44 Is the idea behind this that the reward itself can be used a marker of successful actions? Such if i think left and see the reward the feedback triggers the dopamine response that whatever action i want is correct?

[@rushibhatt1201]

I've been in same debugging process while working on gait classifiers. Super relatable!

[@iAlexSenchea]

Good job. Im hoping that one day you will have a major breakthrough and help revolutionize the world

[@nakulsingh8444]

Awesome 👍 I love it that u take it further

[@dipampatel5622]

You should plug this up with an Arduino or RC Transmitter to drive a car around. Just 4 commands to control. Should be very interesting and fun! :)

[@johanneszwilling]

Dipam Patel 🤪 How about just one: Flappy Bird 🙌🏼

[@marcsmithsonian9773]

And you wanted to turn left but think that your wife is always right and BooM !

[@jalankoe]

Awesome! Tq for doing this

[@koog1000]

Ha! I almost mentioned Nyquist theorem in my last video when you were talking about the sample rate you need. In general, you actually want to sample much faster than the Nyquist rate to have a good representation of your signal. There is actually another equation that describes your measurement error based on your oversampling rate (I forget what that equation is called right now). The general rule of thumb we use in our industry is 10X oversampling rate, and anything less than 5X oversampling is regarded as too inaccurate for measurement purposes. If you look at the jupyter notebook code that you showed in the video, they are also only using data up to ~fs/6, or 6X oversampling ratio.You can see this in the frequency band limit they defined for gamma waves. Long story short, your system with 16 channels at 125Hz is really only good up to frequencies of ~25Hz. I think that in order to train your neural network to understand motor skill movements then you need a higher sample rate. You also should not be using fft data above fs/5.

[@sifiso5055]

Thank you Sentdex😃😃

[@raijinow]

beyond reading up on EEG in general, you could start with validating that your data aquisition returns something other than artifacts by replicating the classic Berger effect (change of power in alpha and beta range with open/closed eyes).

[@DustedAsh3]

I love watching your brain work. Man that is fascinating. I dont mean to be creepy.

[@hmartz]

Wow @sendtex this is finally the year of linux on the desktop ;-)

[@russellcox3699]

Hey sentdex. If things don't work -- try using ffts over time, or something like a stft/spectrogram as an image input to a Conv2d so your network can learn frequency changes over time. If you can do your own ffts on the data, I would also suggest playing with the number of bins and visualize what stfts/spectrograms look like by category. Ideally, you'd be able to spot the differences by eye. If you can, then your net can.

[@stevenfussner2573]

Closing your eyes while thinking may help as it accentuates the voltages of the resting rhythms. I'm not sure if you thinking of the words left/right, but there is a resting rhythm over the motor cortex called the Mu rhythm (an alpha rhythm), just thinking about moving that body part is sufficient to attenuate that rhythm. So thinking of moving the left arm may improve performance in predicting moving left.

[@titusfx]

Your Spanish and Portuguese skills are great!!! 😂 thanks for sharing !!!! 👏👏👏👏

[@ghostindamachine]

Super super cool stuff

[@trolololosh]

These two videos with openBCI are the best videos about it on KZbin, this is awesome. Could you, please, share your work on github)

[@sentdex]

Will all be up by video #3, including training data.

[@memoai7276]

One suggestion I would like to make is that perhaps you could try thinking "left" while eyes closed as the occasional movement of the box to the "right" on the screen might trigger the thought of "right" as the visual stimuli gets processed by the brain.

[@memoai7276]

in any case, this is a very exciting topic. I can't wait a EEG cap for myself :)

[@mockingbird3809]

It's cool to watch you work with these sensors Sentdex. Please try using PyTorch for your Neural Network Building and Processing. Your PyTorch series is really a boost for me in learning it, So Please try using PyTorch for Building Neural Nets and Other cool stuff, because it helps me learn some really amazing ways to process these data inside of a NN in Pytorch. Thanks :)

[@sentdex]

I have plans for pytorch in the near future, but I dont currently plan to work in pytorch here

[@Whakamua.]

What is the reason you don't want to work with PyTorch here? Is it performance or preference or...?

[@qweryedu]

I don't think FFT is going to be enough, because with FFT you can see the components of the waves or the noise in it, etc. But once you filtered that out, FFT is just the componen of the wave you are trying to find; so the problem here is that the wave associated with moving (x-direction) and not doing anything may have the same spectral commponents. I did something similar at school with the P-300 signal.

[@mr.cobalt6695]

If your brain makes frequencies for motor movements, you could use those to figure out left or right, maybe that would work better. Like if you move your arm or wrist left it goes left. If you get that working then you have conformation that you can get the difference between different brain waves and not just the difference between nothing and something. An even better approach might be to use left arm to move left and right arm to move right, since motor movements are controlled by different sides of your brain, and the data coming from the different channels is from different sides of your brain. But yeah if you get that working then you can be more sure you could get other things working.

[@bhaskarsyamal]

Play flappy bird with this thing!!

[@Sharyf]

I was dreaming to have one to test lie detection. BTW It should help to think about moving your l hand\leg during sampling, alternatively thinking about looking r\l. --> By this way motor functions of a brain might have more distinct signal as they are spread in the brain.

[@darektidwell1158]

instead of just thinking left try visualizing something that represents left. Like a road sign with a left turn at the same time as you are thinking left or internally repeating left? Our internal model is very visually driven.

[@Kedriik]

Try using movment of muscles to differentiate right or left; like think to move right hand (but only think dont move) to right and left hand for left. I saw another time documentary about disabled person who used egzoskeleton and this is what they told him

[@snarkyboojum]

Awesome!

[@lautarodapin]

i was really waiting for this vid

[@jcims]

Would be interesting to see if different ways of thinking 'left' impact the signal. One could just be thinking the word to yourself, another could be you imagining pushing the box to the left with your hand, another imagining a fluid on the board flowing in a direction and dragging the box with it (etc etc) Also would be good to try with your eyes closed/display hidden.

[@sentdex]

Have been trying various things like that ;D

[@hernandezurbina]

Great video, Sentdex! Can you tell me what OpenBCI kit are you using? I see that this is part 2, and cannot find part 1. Could you pls tell me where to find it? Cheers!

[@eitanas85]

Thanks a lot for a great content man! did you upload the code to Github already? Would be happy to play with it

[@noisycarlos]

Really cool series. I looked into this, and actually bought two headsets before I heard of OpenBCI. An Emotiv Insight and a Muse. The Emotiv has a demo app and a REST API where you can train actions and trigger stuff. But if you want access to the raw data you have to get their Pro subscription, which is $80/month. The Muse gives you raw data, but it's optimized toward meditation,and their API is not as nice. Both have the issue of not enough channels (which is understandable for their price, and they're not meant for BCI specifically). I ultimately put that project on hold because I realized how much time and effort it would take. As much as I'm curious about it and want to do it. I have another project that I'm focusing on, so I'm excited to follow your adventure and learn by proxy :-)

[@James-ry3by]

What would be the best budget headset to make something neurofeedback related?

[@noisycarlos]

@@James-ry3by Depends on what you want to do, and how deep you want to go. Unfortunately, Muse closed their SDK, so you can't really develop for it on the new bands, as far as I understand it, their focusing more on medical applications. Which is fine, but probably not what you want. That leaves the Emotiv Insight. Which is decent, and actually has one more channel than the Muse. The only problem is that it is $350 or something like that, and that gets you their basic API, which doesn't let you access the raw data, just their predetermined actions. You could still do something useful with that, but if you want to go deeper, and access the raw eeg data, you'd need their PRO subscription, which is $80 a month, or something like that. So, overall, the OpenBCI option is probably the best for tinkering, unless you want to do very simple stuff with the Emotiv. It's definitely more expandable, and while it starts $200 more expensive, you'd end up paying more with the Emotiv Pro subscription after 3 months.

[@James-ry3by]

@@noisycarlos I'm thinking about making a neuro feedback machine to me with meditation and getting into the theta/Flow state. The machines I look at cost £1000-5000 which is out of my budget so I'm looking for alternatives. Do you think the muse is a good option if I can find a way to reverse engineer their sdk?

[@noisycarlos]

@@James-ry3by In that case, the Muse is a pretty good option. There are some third-party apps that give you a lot of data. There is an app called Muse Monitor which does give you raw readings and other data from it. musemonitor.com/ I think you can even export CSVs with the recorded sessions.

[@noisycarlos]

@@James-ry3by also, you might already know this, but if not, check the channel: Cody Rall MD with Techforpsych

[@noisycarlos]

If it makes you feel better, the EMOTIV insight actions in their tester app had a similar accuracy as your model for me. I didn't officially measure it, but it also felt better than random, but not really reliable.

[@sentdex]

Yeah, I expect to do better than that, however :D

[@333jordyg]

Awesome stuff man! I really love your enthusiasm, thanks for making this series. I really want to point you to an issue you seem to not be addressing: What mental function is caused by (or causes) what brain activity? And crucially for an EEG BCI, can that particular mental-function-brain-activity pairing be classified? Thinking the word "left" or "right" is not necessarily caused by (or causes) brain activity that is distinguishable by man or machine learning algorithm SPECIFICALLY using EEG. In the case of imagined speech or inner speech, the brain activity that matters is, perhaps, beyond your reach. By all means, continue trying - if you find a way to make it work, the world would love to know! However, you should be aware that researchers have identified a few certain mental-function-brain-activity pairings that do work well for EEG BCIs. Please look into "motor imagery BCIs." When you imagine a certain body part to be moved, a certain spot on your motor cortex engages a specific brain activity pattern (called Event-Related Desychonization [ERD], in mu and beta bands). Amazingly, a different imagined body part movement will cause the same type of brain activity, but in a different location of the motor cortex. Therefore, imagining right-hand motor imagery will cause a very DISTINGUISHABLE signal from left-hand motor imagery - different electrode channels showing different amounts of ERD you will find in the FFT of the data. Here's a literature review on this topic, I could send more papers if you want: www.ncbi.nlm.nih.gov/pmc/articles/PMC6232268/ Best of luck! I'll be following this!

[@indivarmishra6119]

As for Sheldon(TBBT) Doctor Proton was his inspiration for Science, to me sir you are the Inspiration for programming and education ❤ . Keep doing what you do.!

[@Internetzspacezshipz]

I wonder if thinking “left” and “right” are too similarly related in your brain? Maybe think about other things that aren’t related instead?

[@sentdex]

If necessary I already have data for this. Think none, or think action. I could replace any action thought with left, and none with right. While that would work....itd be super lame IMO

[@Internetzspacezshipz]

@@sentdex Yeah that is true, it would be much more interesting to be able to 'control' something with your mind.

[@MMABeijing]

@@sentdex he meant to say: think about 2.different things, from 2 different worlds,

[@m.t-thoughts8919]

Nice profil pic.😉👌🏼

[@beyhan9191]

I’m thinking about your training data is not correct. I meant when you’re thinking about ‘left’ or ‘right’ there is lots of dirty data(you can’t think only left or right ).Remember it’s looks like mnist dataset. There’s digits ‘7’ but this handwritten ‘7’ is not 100% ‘7’ It’s also looks like ‘1’ maybe 2% so my idea is maybe you can look at left and right data correlations based on channels and you can only consider about negative correlation channels. Because right and left are opposite so this informations must be hidden between channels. It seems you have lots of way to go and needs more preprocessing and data cleaning. Anyway good job thanks for sharing your experiences with us!

[@ayeoh47]

I agree, the data cannot be accurate without proper cleaning/pulling of what wave function(s) demonstrates proper LEFT and RIGHT.

[@UnderTheTutelage]

Have you tried this experiment with your friends and family to see if there is a difference in the way their brains 'think' a class label? That'd be interesting if you can somehow quantify a similarity or dissimilarity measure between persons and create a relative scoring.

[@JustTechGuyThings]

@sentdex, the example has daisy set to False... set it to True and I think your 63hz should be cleaned up.

[@MrVaunorage]

Please write a blog about it !

[@wktodd]

There is a possibility that the headset is picking up eye motions , so if the spot moves and your eyes follow it it will re-enforce that motion regardless of your thoughts. you may need to change the visual stimuli so that your eyes do not move (e.g an arrow or a turning disc)

[@sentdex]

Possibly, but the training data contains a LOT of samples where the square is actually going in the wrong direction. At the moment, about half of the samples are FFT data of some action where the box is actually going the wrong way, so I don't think that's gonna be an issue. I like the way you're thinking though :)

[@wktodd]

@@sentdex But, if you're thinking is causing your eye movements , as your willing it to move (a typical human gesture ) that will also add errors to the data (assuming you are not trying to make an eye motion detector) .

[@pasquale-s5g]

have you ever made a tutorial on text generation?

[@CyberAnalyzer]

a sampling frequency 2 times greater the data frequency can still bring to ghost frequencies, that you don't want! with 10 time greater you are on the safe side!

[@danielalejandrofigueroaari9240]

Hi, nice video. What if you put some music and save the data to make songs recomendations based in that data.

[@lizardy2867]

I wonder if a keypress alongside the thought could help with training. At least that way you'd be dumping data you know absolutely correlates with a specific direction. (also you won't have to constantly edit the script to change the direction you're trying to think)

[@sentdex]

Key presses will generate unwanted EMG noise

[@Akshatgiri]

Imagine training a model with 26 output neurons so you can type with this.

[@user-qw2tt6su7p]

@wise guy in all, he's mostly just tinkering and having fun, but with the goal of using a CNN and possibly an rnn to classify 2 or 3 outputs to left, right, and mabey none. In a previous video, it looks like he made a simple pygame script that moves a box right or left. Where inputs/input will be the eeg data.

[@PyMoondra]

Are you thinking left or trying to activate the left part of you brain? Fluctuations seem to be active with fine movements.

[@SeucheAchat]

I would be interested in knowing how much time you spend on this topic between the Two videos

[@abangilhaam7286]

how to make brain commands like voice commands?

[@kevinfutero7166]

Nice, any idea of how much does it cost this EEG kit?

[@marcsmithsonian9773]

Ok. You have long toad ahead.. I can tell you how to make it works. Firts hint is to decorelate data. Try data reducer PCA works poorly Kochonen map is better. Preprocessing snapshots should be there done with gaussian smoothing etc, as you feed alot if noise. Values also can be run with some fastvreacting cycle length measurment like lomb periodogram, Gortzel, or Maximum entropy spectral analysis. To extrapolate cycle length , magnitude and signal to noise phase etx.. btw good work ! And good Luck.

[@MufasaToday]

I bet you tried to think left and right when he was doing it.

[@wpirkl]

Cool Video! Maybe push left out right key (our none) to let it know what you think?

[@sentdex]

Pressing the key will generate EMG noise that would otherwise not be there in practice

[@wpirkl]

@@sentdex good point! In fact you have to keep super focused to get things working... What are those high pulses you get from time to time?

[@KarlBonenberger]

I'm guessing there's a dataset directory on his drive named .bigboob and .littleboob LOL.. ;-) Luv it..

[@firstnamelastname3106]

why rnn/lstm can be "worse" than conv when it comes to sequential data ?? never expected that to hear.

[@Imdeepmind]

What happened with your pytorch series?? Do you have any plan to upload more videos on pytorch?

[@sentdex]

yes

[@Imdeepmind]

@@sentdex when? I'm desparately waiting for the tutorials

[@parkerdinkins5541]

The x-axis on your system is 1200 while the example is at 600

[@chaostoorder8736]

Have you thought about using more differentiating thoughts? What comes to mind is a scene from Harry Potter where he must think of an exceptionally happy thought to conjure a Patronus. Thinking "left" or "right" literally might be too close to one another in he frequency domain to reliably measure. Therefore I postulate using drastically different thoughts to MEAN "left" or "right" may lead to better results. (i.e. Think about a really cringy moment to go left and think about a really proud moment to go right). I expect that using thoughts that light up totally separate parts of the brain is the way to go.

[@manikanta3977]

Hello hai i am want to learn machine learning from scratch and i don't know where to start and how to start and i am new to your channel i seen your playlists it was good please guide me ...i know python basics ..... please let me know how to start ...i want to learn data science so machine learning should be main thing .... please reply...... please..... what tools actually required for data science and modules ....some use bigdata Hadoop spark etc some use tensorflow pandas numpy etc ... what is reason

[@ahmadanis9930]

❤️

[@Traincraft101]

Will you ever do a tutorial on the dc-tts model?

[@dipampatel5622]

Yes, even I'm waiting for it.

[@sentdex]

Possibly. It'd be pretty in depth, take a ton of work... all for an extremely niche topic that will likely perform quite poorly, so we'll see. If I get bored enough, maybe :P

[@asdfghj402]

Brain-driving cars in GTA V when

[@sentdex]

Immediately after I train a decent model !

[@bruhaspati560]

Play some games using this stuff That would actually be cool!!

[@sentdex]

I plan to hook it up to GTA V if I can get directions and stuff worked out.

[@robertobomfin3787]

About the fluctuation that you see. Maybe it is not the sampling rate itself, but rather the rate that your samples are transmitted to your PC. If this is the case, then there should no issue. Also, regarding the FFT, I expect that your system should be able to work without performing the FFT, that is, with the samples in time doman. The reason is that the FFT is nothing but a change of domain. You get no additional information by doing that. Maybe it is worth giving it a try in order to decrease processing time.

[@sentdex]

FFT is a method of preprocessing an otherwise large, and quite noisy, set of data.

[@robertobomfin3787]

In addition, I would also expect that filtering the 60 Hz noise with cause no additional gain in performance. I think your system would interpret it as some useless information because it is always there, but would cause no harm. But I am just speculating =D.

[@robertobomfin3787]

@@sentdex I mean, the FFT itself if just changing the domain. From time to freq or freq to time. That is true that you can do things like filtering, but that is something beyond the FFT. In your case, if filtering your signal for whatever reason in frequency domain is really necessary for some reason, then it seems that the FFT is unavoidable. On the other hand, if this is just an operation you do and nothing else, then I assume it would make no difference. I highlight though that I am not familiar with deep learning and it is possible that I am missing something.

[@robertobomfin3787]

Still about the 60 Hz signal. Theoretically, I even think that filtering it makes you to lose useful information, and therefore decrease the performance a little bit. The reason is that whatever filtering process you use, it will not perfectly doing it. This means that you might delete a bit of useful data there. However, I believe this loss should be negligible. I am just brainstorming ^^.

[@ahmetakil787]

finally

[@vincentdavis4915]

Maybe you have to wait until the framerate gets up to speed before training??

[@whoisabishag3433]

"A DisAgreeable Basement Dweller ... 04:10 ... "!"

[@12mkamran]

Maybe try doing some meditation and look at the graphs?

[@stardust7405]

where is part 1???? :(

[@alexandremarcotte7368]

I started a GUI in Python to visualize data Live from OpenBci if someone wants to branch from it: github.com/AlexandreMarcotte/PolyCortex_Gui

[@rajeshSimpleton]

what actually happened. During the *Left* cycle: Harrison: Move Left Don't move *Right* Move Left Damn. Stop moving *Right* I though Left. Not *Right* Neural Net: You said *Right* ? Got it.

[@fishermikoaj6578]

Maybe you could use activity per electrode, my guess is left side of brain will be more active when thinking "right". In general in brain there are regions specialized in task like speech recognition, movement etc. Did you think the word "right"/"left" or about idea of right/left like: movement of right arm leg, right side of the picture etc?

[@sentdex]

That's already what's happening. The FFT is a graph of all 16 electrodes. The convnet is a 16-channel 1D convnet, so the data is there with regards to left/right for the model to recognize.

[@abdulrahimkhan9285]

Wait you also switched to linux from windows? I was having issue on installing spacy it tells me to download microsoft visual 2010 c++ 14.0 But I did as it said from the official website Then it accepts Microsoft visual 2019 And said "----" file missing I forgot the file name So I switched to ubuntu and it worked fine Damn, windows sucks

[@FuZZbaLLbee]

I was thinking, left and right are both abstract terms with a lot of the same relationships in our mind. Wouldn’t it work better to think about fe “race driver” and “grandma” to go left or right? Provided that your granny isn’t a race driver of course.

[@olee_7277]

Try a GAN :)

[@johanneszwilling]

🤓Wonder if this requires meditation at some point. Thoughts are much harder to control than physical actions.

[@ahmednayeem4849]

so true

[@Vladeeer]

Do i need to shave my head if i want to use this headset ?

[@sentdex]

I have some hair and it works. They are dry pokey electrodes. The more hair you have to get in the way, the more annoying it might be getting started and connected, but otherwise no problem.

[@MrMultiMediat0r]

8:15 - didntlookright.pdf XD XD XD XD

[@subhraneelmazumder9079]

Why does he use Brave browser?

[@asdf-ef8if]

Why are you so cool?

[@businassman]

If Charley controls you with this thing on your head, let us know

[@phillipotey9736]

What happened?

[@Gavinnnnnnnnnnnnnnn]

what's the benefit of using ubuntu for this?

[@sentdex]

Like I said in the video, I really just used it because I was already dual booting, and Windows was failing me :P

[@FuZZbaLLbee]

“If I am not thinking left or right is very accurate, and that probably because nothing is going on” That sounds worrying :-p

[@sentdex]

Or I am a meditation expert, capable of achieving nothingness.

[@vishwasnarayan]

which OS are you using???

[@sifiso5055]

It looks like Ubuntu

[@shreyassable4981]

Ubuntu 18.04 I guess

[@sentdex]

Ubuntu

[@vishwasnarayan]

@@sentdex is the ubuntu 18.04 or the 16 because you had used 16.04 for kivy

[@vishwasnarayan]

my pc is old what should be the OS for it?

[@taba1950]

I'm not any sort of an expert on this but I think you shouldn't be discarding half the data, it Should be sampled down or that what I think

[@alexandremarcil498]

Can you share your code please!

[@sentdex]

Part 3 will share code Aaaaaaand training data.

[@mathematicalninja2756]

Ayy

[@FuZZbaLLbee]

Isn’t a convent for classification and RNN for predicting t+1 ?

[@sentdex]

Historically that's been fairly true, but convnets are outperforming RNNs for sequential data, especially convnet + attention. More on this later in the series possibly.

[@genugzocken]

Hi sentdex, hi guys, I am totally stoked! Would anybody or master sentdex himself be so kind an tell me what model is used in this series? Is it his one shop.openbci.com/products/d-i-y-neurotechnologists-starter-kit?variant=16117684338760 for 1495 USD? What components needed to be 3d printed? Maybe I missed that in the last episode? Thank you very much!