Machine Learning with No Code

[@realamanpriyanshu]

you are the best. I am starting new from all the motivation i have gotten from your videos. Thanks!

[@mustafabohra2070]

You are the reason for the inspiration of million AI enthusiasts on the planet!

[@wolfisraging]

Machine learning without code is possible, but without Siraj it would be boring. Lots of love

[@nigeriaconcerns]

gbam

[@bloodaid]

Siraj is the entity I pray to before I go to sleep every night

[@wolfisraging]

@@bloodaid 🍺

[@nigeriaconcerns]

@@bloodaid that is extreme idolatry

[@bloodaid]

@@nigeriaconcerns He provides me knowledge that could change my life. I owe him an organ. And I'm not talking about the instrument.

[@evgeniymamchenko9582]

Another way to try transfer learning for image classification without writing any code is an Android app called Pocket AutoML. It trains a model right on your phone without sending your photos to some "cloud" so it can even work offline. Disclosure: I made this app.

[@digital_down]

One day it will be as simple as, “hey Siri, make my own personal image compositing software in a node based architecture”. “Ok, here’s the software you requested”. Keep telling Siri what you want to add and subtract from the software until your needs are met.

[@milton5417]

Digital Down Intriguing.

[@mangomaster2005]

That's basically how Tony Stark "programs" his AIs like Friday and Jarvis ... Or Star Trek did their interactions with "Computer". Reality hitting SciFi again.

[@arianphilips5777]

so for doing that tjere has to be a mind behind imagine the paya

[@cedricvillani8502]

Uh well learn Wolfram language, the top 50 all use a Wolfram product, have been for years. Siri is from Wolfram Alpha (a specialized build for Apple ) AI models that get released are either out of date or from edu, but even then a paper won't be published publicly if it is ground breaking and gets that department a nice “Grant” I'm pretty sure it's still high 80% that's under heavy NDA and proprietary. The rest you see, or we see. Being a graduate in Electronics Engineering (not electrical lol) I have been apart of iEEE since 2001, there are projects there that are incredible and with the correct membership and degree you can even join in on a project. I think they have a open to the public membership that's fairly reasonable price-wise, and you can join various general groups that drill down pretty deep but that's when it gets expensive, but if your in EE CIS IT..etc they send out a magazine called Spectrum and most of each issue is JOBS available, like a lot so check it out. I've been doing “lightwave technology” for past 5 years, it's definitely more than that but I can say the general area of focus.

[@NickJohnsonHill]

H2O Flow is also really good for codeless Machine Learning. They also have Auto ML to have a leaderboard of any algorithms and hyperparameters against each other. Similar to Data Robot bit without the feature engineering.

[@prashantgoel1659]

can i use it to build a ai which can generate snack game or flappy bird?

[@RAOUFTV16]

Life is a binary machine learning classification problem ( happy | unhappy ) ! We start from a random vector w that we did not choose W = [ Country , Parents , Culture , Financial situation ... ] The loss function is the MSE the dataset is life experiences Goal : best (happy | unhappy ) predictor for further situations and experiences Optimization : Mini-Batch Gradient Descent

[@innocentiuslacrim2290]

This is great that we are heading towards this direction. There will always be a place for fine tuned models that reach for the best possible accuracy and efficiency, but making modelling and deploying easier to use we will have a lot more applications actually made and deployed. There are plenty of businesses that either lack their dedicated machine learning specialists or have them, but they are so busy that they cannot address each possible problem. There are also a lot of non-profits, researchers and private individuals that could use these techniques but they lack the deep skills that have been required so far to use them.

[@earthnote]

this channel deserve more subscribers and likes

[@PieroMolino]

I’m noticing this video a couple weeks too late, but thank you so much Siraj for this video and the kind words you spent for my brainchild Ludwig, much appreciated!

[@OnlyGoodJawn]

You’re so hood Siraj. I friggin love it my G.

[@TheEtsgp1]

Awesome video LUDWIG rocks im using it to read samples of random video datasets with no audio and test sets with audio to read lips and output what people are (or could be) saying! and to answer your question about the favorite tool Im all about JUPYTER notebooks man it does everything! im surprised you haven't covered the usage of it in detail? unless I possibly missed it? that, and virtualenv are just the beez honey for me dude!

[@marat61]

Hi Siraj, can I do machine learning without knowing alphabet?

[@SirajRaval]

not yet

[@ayubomaraden867]

You can make AI to teach you alphapet

[@marat61]

@@SirajRaval "yet"?

[@danielbadra9447]

what is alphabet? I'm considering learning ML, but never heard of this.

[@danielbadra9447]

or do you mean like... a literal alphabet?

[@americanpsyop1364]

Damn Siraj You still explain this video in Python Format

[@ponpokofamily]

Thanks for video wrapping trend. Very happy to see that trend is going code free direction. I worked on enterprise package software called NeuralWorks made by NeuralWare around 1993. It was code free visual tool but was very expensive. At the point of Yr1993, it costed around $30K.

[@xoxo-sf1zg]

I have fallen in love with your video title. I hope you don't cheat me.

[@edoardog1899]

Great video Siraj!

[@kiranjotsingh4040]

great work bro, keep it up, u r a saviour, recap of video in last is a great move

[@mathuff5]

Great video as usual. Thank you for sharing!

[@lopezco]

I like Dataiku for data exploration. You can code if you like it, you can "drag and drop" if you like it, but also it has a simple HMI for automatic model training and selection. I'm not saying that it is the best tool there is, but it is nice.

[@username42]

so which one is indeed free for usage ? i think only azure has some free hours for students and deep cognition has a free desktop version but not datarobit feely. and also what about the knime blueprint?

[@maheswarisrinivasan8391]

Amazing! Lots of new info. Thanks

[@ddrussianinja]

4:43 Uhh, and how exactly is one supposed to just download the Cornell Movie-Dialogs Corpus as a CSV? They only provide it in raw format and your source code link doesn't provide it (or the colab environment you used, for that matter). Yes, we can put together code to convert the raw data into CSV, but then we're not doing machine learning with no code, are we?

[@jude3046]

Yeah I'm trying to follow along as well and we're missing a step here for sure

[@ddrussianinja]

@@jude3046 I played around with this a bit yesterday. I found a DB version of the corpus and used a simple query to output it to a CSV file in the format shown in the video (you can find the format info in the Ludwig Examples section under "Chit-Chat Dialogue Modeling through Sequence2Sequence", which is also where they have the yaml file written out) and got Ludwig to load it all in a Colab GPU environment, but then I started getting out-of-memory errors, so I had to disable the TF_FORCE_GPU_ALLOW_GROWTH environment variable before running it (the command is "env TF_FORCE_GPU_ALLOW_GROWTH = false"). Colab still warns me when I run it that it's running low on GPU memory, but if you ignore the warning, it'll work fine. But even once it starts running properly, it takes a very long time to train in the Colab environment, probably around 10 hours or so, and Colab sessions recycle completely after about 12 hours, so make sure you're around when it completes or you'll probably just lose all the work and never get to test it. The CSV version of the Corpus I made is here (for the next 30 days) if you want it: ufile.io/eysat

[@jude3046]

@@ddrussianinja nice work Pat! I'm already trying Ludwig on my own projects but thx for that info on Google Colab. I'm pretty new at this so I'm sticking with local deployment for now, I can only handle so much complexity at a time haha.

[@theundaunted5481]

Thanks

[@Lazy84.20]

When I try to run this, I get an error that says "KeyError: 'class_weights'". Any idea why this is?

[@M.G.R...]

*AI without codeing is possible, but without Siraj is impossible*

[@SDC-02]

Great news! Thanks for the update.

[@herm41]

RapidMiner also very good no code machine learning tools. RapidMiner named a Leader in the Gartner Magic Quadrant for Data Science and Machine Learning Platforms

[@vencislavkirilov6530]

Greetings Mr. Rival! There is a very interesting free tool called Perceptilabs. Unfortunately, there are a few unknowns, at least for me, related to data preparation. Otherwise, it seems to me that it is the perfect tool for people who have little or no programming skills. And yes Mr. Rival, If you can help and explain some things with this program, I think it will be interesting and useful for all! Thank you and good luck!

[@vigneshamudha821]

can you explain hebbian learning and also difference bet hebbian and gradient descent

[@michaeloconnell1138]

Can these models be exported and converted to work with tensorflowjs?

[@ChrisSimokat]

Good video. Bringing up DataRobot definitely makes sense, and has some powerful features with a slick interface, but no love for H2O's Driverless AI? They play in similar spaces, but DAI runs on GPU's. Regular H2O also contains AutoML and through the Flow user interface can be setup through basically filling out a form. No coding required.

[@rickharold69]

So bloody awesome. ! Thx for video!

[@bkstyles5769]

hello there, why it seems that the IBM Watson studio is more simple to train model by just feeding the data such as image etc to trained model-like image classifier etc, has any thought on it? thanks

[@jameswait2934]

Thank you Siraj! Just about to download Ludwig...

[@lawrencewong3367]

Your example references the Cornell Movie dialogs but the files I downloaded are not parsed in proper dialog sequences, unless I do some pre-processing to match them. The example model_definition.yaml in Ludwig website just have user1 as input and user2 as output. Where can I get the chitchat.csv as defined with user1,user2 format without having to pre-process the movie_lines.txt?

[@AghaKhan9]

Great work siraj

[@saketojha494]

Great fan of ur vidoes Will surely Wrk on this Ludwig....

[@dr.tanzilasaba9847]

That's amzing. I hope you provide us with a way to use these platforms.

[@geronimotutusaus]

Knime and Rapidminer are far better than Azure ML. That would be my recommendation. Love the democratisation of these technologies.

[@masterjiggle5728]

Liked your video I only want to ask one ques, from where do you get suck websites?

[@chrisj21b]

Siraj, it seems that deep cognition has done everything you suggested as a start up opportunity, what do you thing is missing about drag and drop implementations in frameworks as tensorflow?

[@kunalr_ai]

Thanks for the episode

[@username42]

where is the colab notebook you showed in the video, bro ?

Siraj, as usual Your essence is enlightening my AI concerns. Don't You think that Ludwig could introduce the commercial bias into the bias we all know ? Yours PEF

[@econrith]

How do i give the model only the music and sounds i want it to train on? in other words how do i submit to what ever would make of the music I love .csv files? so that the models can understand it? I understand that there are music data-sets available as already prepared but what could prepare my own selection?I 've signed up for your course.

[@ashaharika9565]

can you please upload a video regarding what is sparse represented coefficient matrix and why we use it? please,,

[@Marc42]

I'm mostly a KNIME user myself (@KNIMETV), but Ludwig seems like the kind of simple fun to add into the mix!

[@ulysseskoudou962]

I love you siraj like a clean dataset, you and I are just in perfect symbiosis !

[@orozcodavid101]

Can we get a drag and drop surgery kit too, maybe some DIY lasik ? Why not make everything automatic !

[@omarlopezrincon]

Hi Siraj... do you think it is possible to create a VAE without any library ??? like in the Math of Intelligence ? Greets

[@Alex-xc3ki]

That's awesome. I hope you provide us with a way to use these platforms.

[@kartikathale1483]

How about splunk machine learning toolkit. Is it also a no code tool. I need know about this tool. Thanks!

[@ingusmant]

What you mentioned about drag & drop AI applies to any d&d programming really, people think the problem with code is that's all text but the real bar is understanding the code not turning it into pretty icons BTW the mind expansion meme should be the other way around

[@vishzp]

Awesome! Can you do a video about how AI has evolved in the field of game development like procedural generation.

[@thee2296]

Have you ever used Knime?

[@samuelthudi9181]

Came for the memes, stayed for the content. #respect

[@PrateekGole]

So, should one consider a career in data-science, ML and AI? If yes, what skills (and approach) should one focus on while learning AI?

[@jamesbarker6373]

Dammit, I knew this would happen. So, should I stop learning Tensorflow?

[@FuZZbaLLbee]

James Jameson Yes! Knowing how it works is totally useless :-p

[@jamesbarker6373]

@@FuZZbaLLbee My life is a lie, lol Well, the past 2 months have been anyway Thx4dreply

[@ingusmant]

Stop learning everything! Next week there's some new crap coming out that's exactly the same but the logo is purple! PURPLE! YOUR CAREER IS OVER!

[@SirajRaval]

no keep learning, these tools aren't nearly as effective as code.

[@digital_down]

One day... but not yet. It will still take many years until Tensorflow is obsolete. It’s still very useful to learn.

[@lamkasdev9983]

9:32 ok, but programming line by line is still faster and better than drag & drop

[@GS-gi9bc]

WIth business intelligence and an interpretive knowledge based engine, the software can generate the code, line by line. Then it can be packaged to create the process function by function, then business process by business process. It all depends on the ability of the knowledge base engine to know how to be trained. I'm looking for a venture capitalist to partner with my software to roll out business based knowledgeware, targeting specific industries. Yes, it does mean a lot of line by line coders will have to take up some new skills to make a living. [e-mail] if you are looking to sponsor such a project.

[@diegoantoniorosariopalomin2206]

Siraj , i have a question, can neural odes be trained evolution through strategies through safe mutations ?

[@diegoantoniorosariopalomin2206]

Also can shufflenet and or deformable convnet ( most important paper you haven't covered ) be adapted into neural odes

[@georgemavimbela]

Siraj you are the best

[@edfernandez322]

Hi Siraj, what's your take on other ML tools/platforms such us Bigml.com h2o.ai or dataiku? Thanks

[@vaibhavacharya9428]

So basically all my knowledge acquired just turned useless...Drag & Drop...

[@pakpoomtivarkornkit8525]

That's what I'm thinking...

[@jakecostanza802]

Eventually that's what will happen to all human knowledge, your just one step ahead. :)

[@KenJee_ds]

Great video Siraj! I think AI for the masses is a great concept, but could also be quite concerning. When consumers use any form of advanced math to make decisions without an understanding of the underlying concepts, they are setting themselves up for failure. We get into a grey area where people say "the computer said it so it must be right", they don't realize that there could be problems with the data quality or the underlying model decisions that the AI is making.

[@alexanderwestberg795]

Hi Ken, for someone who is new to ML, do you think it's necessary to learn the concepts that ML models are based on in order to learn and analyze data in good way?

[@KenJee_ds]

@@alexanderwestberg795 I don't think that you should have to be able to code all the algorithms from scratch or anything, but you should have a high level understanding of what makes them tick. It is important to know: what makes one model different than another, if you need to have normalized or scaled data, and if the model makes any other assumptions that need to be taken into consideration.

[@alexanderwestberg795]

​@@KenJee_ds​Sounds very good, thanks for the tip!

[@KenJee_ds]

@@alexanderwestberg795 No problem! Thanks for subscribing to my channel!

[@konasatsuma]

I'd say consumers using tools that run on 'advanced math' is no more concerning than data scientists teaching on KZbin :)

[@AC-ep8kl]

You forgot the link to the Azure video

[@yasertake5983]

I did try installing Ludwig from cmd prompt but I am facing error while importing, can you make a video with the installation and using it with jupyter notebook

[@fauzirizal4527]

Thanks for the info, really helps!

[@clericuzio4924]

Thanks a lot Siraj Raval !!!

[@RidingWithGerdas]

KNIME is legit tool for Data Science and for Data Mining! It's open source, you can use Python and R codes inside of it, easy to use SQL connections also. it's awesome

[@toddbascombe8327]

Hi Siraj, should I just stop studying ML programming and just use Ludwig instead?

[@jupiejupe]

Stock Price Prediction with Ludwig please.

[@denismerigold486]

what computer configuration is needed to practice machine and (deep) learning???

[@razataggarwal7365]

You just need a strong internet connection to access Google collaboratory or cloud which provide computing resources for free.

[@denismerigold486]

Thanks

[@hasantarek6521]

Keep going brother

[@crm3q]

Hey, how can I relate coding and technology to being in medicine( medical Dr)? I currently am learning coding and love it but I still want to complete med school. How do these relate? Is there a job combining both fields? Thanks for any advice.

[@amuhlongwane5714]

Hi Siraj, Good work as always. How is this different to AutoKeras and Google’s AutoML?

[@anujlahoty8022]

I really loved your video.. Could you make a small Demo on this please??

[@hsyhockey30]

One I have enjoyed is the KNIME platform.

[@davidbonn8740]

Siraj, Thanks for the great video. Tools like this are nice and obviously will become the way applications using deep learning are created in the future. However, this doesn't make the job *easy* or *simple*. My limited experience so far is that 90 percent of the work in any deep learning project is scheming furiously to build a bigger and better dataset. And I am convinced that for the vast majority of possible applications the datasets you'd need don't yet exist, or if they do exist they are awful or fundamentally broken in some way that makes them useless for an actual application that people would use. I suspect that someone is going to make a Great American Fortune developing curation tools for datasets.

[@obinator9065]

C++ isn't hard. Code can be quite trivial if chosen the right resources. I mean you could develop a gui with Windows for example and expect about 1000 lines for a medium application, or you could use Qt and obtain a Java similar experience, with way less code. C++ gets really verbose when you optimize it(simd, pointers, ...)

[@vikingxd8644]

AI would know that you showed a Puffin not a Penguin.

[@Nv2121]

is learning computer science with an ML specialization redundant now?

[@hydrogen8898]

Rapidminer..how abt that? @siraj

[@aidenstill7179]

Do a lesson on creating your own deep learning framework. please or tell me where it can be studied

[@IdiotDeveloper]

Thanks for the awesome video

[@jude3046]

Hey Siraj, sweet video! I noticed Ludwig on reddit and thought you might do a video! Having trouble installing despite following the Github instructions closely...created a new virtualenv just for Ludwig...then pip install ludwig and it went great all the way until: Could not find a version that satisfies the requirement tensorflow>=1.12 (from ludwig)(from versions: ) No matching distribution found for tensorflow>=1.12 (from ludwig) In the Ludwig github it says we need tensorflow>=1.13.1 yet the latest release from tensorflow is 1.12 No idea how to get this working, any thoughts? EDIT: I'm dumb...still can't use Python 3.7, need to setup your virtualenv with Python 3.6

[@ayesh4646]

Is this similar to Auto Keras?

[@passionate_gaming4629]

Hey brother, I need some help. I am going to do a project on food(vegetables,fruits,etc) recognition system using AI. I have to complete my project in 3 months. I know only basics of python. What should i learn? What path(or steps) should i follow? What are the frameworks and algorithms should I use? Please help me.

[@MadhurjyaBora]

can you please do a lstm model for classification? for example heart disease classification. please help or else I'll fail my last semester

[@yvetteg]

I do not know how to code but u am interested. Do you have to be good at math to get into ML?

[@arkasaha4412]

So do you think if people are able to build AI without coding knowledge would render the field just too trivial? What about those who want to make a career it AI and are learning new frameworks and languages?

[@Kafuggi]

Am also asking myself the same question.

[@arkasaha4412]

No one else seems to be bothered.

[@Iamwolf134]

Well, to be honest, Ludwig doesn't eliminate coding entirely. If anything, I can see Ludwig being used for extensive and quick beta testing so that human coders can refine their code. I can also see Ludwig being used for more mundane coding while Humans pour all of their effort into more special code.

[@NeyTorres]

I downloaded the corpus. no csv file on sight.

[@paras-aiandstartups5457]

You are a true inspiration , you handle situations too easily 🙂🙂 I aspire you the most ,I don't have words

And if a want to start to code, how do I start ?

[@bazareta6936]

So know I should n’t learn how to code and all that of pytorch, or Tensorflow ??!?

[@electronresonator8882]

can I use Ludwig to create an AI that similar to AlphaGo?

[@kushalgupta9613]

F

[@FearzEnemy]

insta office reference, good shit

[@antonykevin2876]

Nice one.. you missed Google cloud visualisation

[@kalashchangal170]

I bet more than half people want to do ML without code

[@xoxo-sf1zg]

Without math*

[@bazareta6936]

Sure, If you have the simple and direct option

[@alfinnurhalim2507]

I Think it's good in productivity. I Mean sometimes we need something that fast and reliable program to do our job.

[@mohanpremathilake915]

Thanks Siraj :)