Build a Chatbot - ML for Hackers #6

[@jakewaitze5104]

Thanks for another great video! I was actually just playing with Karpathy's implementation of a character-based NN the other day. I can't wait for more videos, and you really ought to have lots more subscribers sooner or later because your channel is a gem in the vast YT space.

[@SirajRaval]

Thanks Jake, I'm going to keep pushing content and hopefully they will come :)

[@jordanjohnson2770]

This is awesome. Super good hosting - smart and not annoying. Nice work!

[@SirajRaval]

+Jordan Johnson thanks Jordan!

[@BuckeyeStormsProductions]

I talk with a lot of real humans on a daily basis. Some of them don't even pass the Turing test.

[@camrongodbout1]

Nice combination of technical detail with high level abstraction

[@SirajRaval]

thanks a lot :)

[@rameshmaddali6208]

I am with Siraj - on this point - Govts get prepared for basic income for your people - get ready - Dear people get ready to enjoy your 24/7 time what ever you want. No more saying as "god i don't have time" - you are freed from work and intensive labour jobs. all jobs will be done by machines soon. I am glad i am part of it making this happen. Hope we will satisfy our passion. I want to be a painter but no time is a reason for this desire to accomplish. But one day if the machines are working on behalf of me i can pursue my passion of being a painter. Wooowwwww

[@SirajRaval]

exactly! agreed Ramesh

[@rameshmaddali6208]

+Sirajology hi siraj i am glad to get reply from you - most of my friends who are not aware about whats going to happen They can not able to believe one day driverless cars, job automation, general AI enabled machines will be with us kind of helping us. They can not believe with me. With your reply, Now i have proof on AI and bots stuff. Simply third person judgement haha

[@rameshmaddali6208]

It seems you are good enough for doing crazy ML stuff by seeing your vedios i can say 95% confidence you can do it - i would like to appreciate your efforts you have put in for achieving it. I know its not easy stuff but can be achievable with right guidance and consistent effort. I have one idea which will help break language barriers in the world and India especially. Instant context based language translator which will bring good bonding and closeness among the diversified country like India or World. Better way to communicate each other seamlessly. Its needed for the world which can be possible with NLP and stuff. It will help atleast the next generation of the world.

[@simonstrandgaard5503]

AI interests me and I have been coding smaller projects. I like the speed of your video. Technical talk that is easy to understand. Thank you Sirajology.

[@SirajRaval]

Thanks so much! I'll keep going

[@elitafaye]

Really appreciate your enthusiasm in these videos!

[@SirajRaval]

I'm flattered, thanks!

[@sushantahuja9067]

This almost feels like the guide books we use to have while solving ncert books in India... Straight answers without any thought development!

[@robienob89]

hey Siraj, I really think your videos are great... probably the friendliest AI channel on KZbin. Had a question, do you know a Python library that can generate conversations based on key words? Also do you recommend any topic identifier libraries( topics based on sentences). cheers and keep it up

[@SirajRaval]

thank you so much :) hmm try this pypi.python.org/pypi/ChatterBot

[@FloreVEVO]

>yoyo whats the time-izzle my nizzle I lost it there

[@__A.J__]

yo I died there man :v

[@sri9222]

You're brilliant Siraj! Became a big fan of you

[@SirajRaval]

Thanks Srinath!

[@feihcsim7045]

entertaining and informative.. you're killing it man

[@SirajRaval]

Thanks Brian!

[@SanjinDedic]

Dude I just committed myself to watching all your vids. Awesome work, point of taste... I think your vids would work better with more time on the code itself. Most people (me included) are not at a level where the code is a cakewalk once they understand the concept!

[@liwang9682]

First time on your video, awesome!

[@SirajRaval]

Thanks so much!

[@nanditagoel1983]

Hi Siraj, I like the definitive topics you explain. But I would recommend you to please slow it down in explaining stuff, especially for the newbies. I tried decreasing the speed of the video but it then sucks. Thanks!

[@leemelbourne3297]

Awesome video. Very funny and also insightful.

[@leemelbourne3297]

Would be great to have a Python example.

[@MintoIssac]

You are awesome man.. this video answered most of the questions I had

[@SirajRaval]

thanks Minto

[@ezez4205]

How does one decide how many hidden layers to use? Same, how do you decide how many neurons should be on those hidden layers? Also, how do you know if you should connect all neurons to all other neurons, connect all one layer to the all next layer, connect them randomly, etc, etc etc??? Good videos by the way. Im studying CE and this helped me a lot!

[@SirajRaval]

hidden layers -- the more the better (but you need the equivalent computing power to train them all) number of neurons -- same deal. knowing where to connect neurons -- depends on the use-case. feedforward, recurrent, etc neural nets all have different neuron connection patterns and they produce different results. think of an application first, then build from there. you'll discover the optimal ML tools for the job organically

[@ghctoob]

Brilliant and info packed! Where'd you get the background on so many libraries and theories? I guess there are no call centers in Switzerland!

[@SirajRaval]

Ha lot of time spent browsing the internet and trying different code repositories. I learn best by doing. Also I don't understand your last statement. Thanks for watching!

[@ghctoob]

+Sirajology Switzerland just rejected universal basic income, a ref to your comment in the video. would love to hear more about your discovery process. mine is scattershot at best!

[@SirajRaval]

Ah, good point! My discovery process is like -- browse hackernews, machine learning subreddits, futurology subreddit. I learn so much from reddit, its crazy. People call reddit a time wasting site, but I get so much value from it. Then once i understand what the most bleeding edge libraries are, i go search for them on github

[@emilianstoimenov737]

Fresh, keep it up, love your energy and enthusiasm. Subscribed.

[@SirajRaval]

Thanks Emilian will do!

[@aakashshrestha8120]

seriously man you are genious.how you make yourself to learn this much things ??

[@doubtunites168]

dude tay tweets was so ahead of the curve 5:06 hashtag next level shit

[@mefaltan5paraelplan]

What you recommend for text(strings) inputs for NN? That is to say text transform to number inputs for NN.

[@SirajRaval]

word2vec

[@ShadowMystic7]

I've only been subscribed for about a week, but if you keep honing your delivery, you're going to end up being the Bill Nye of Machine Learning. Perhaps a little more PG-13 though.

[@SirajRaval]

thanks Jonathan. Haha yeah i've gotta to try to be more PG. its hard

[@ehsanmo1367]

Nice! Keep up the great work :)

[@SirajRaval]

Thanks for the encouragement!

[@cortexai6541]

Really awesome stuff bro!!

[@SirajRaval]

thanks!!

[@castm647]

Will there ever be a model powerful enough to filter out all the noise in your shirt

[@SirajRaval]

hahaha! Thank you for the feedback i'll cut down on the shirt noise. XD

[@aaditya2952]

Checkout my python program where microsoft advace AI chatbot called "Ruuh" is chatting with other pandora AI , Two intelligent AI chatting with each other : kzbin.info/www/bejne/jpTchKF8jdWSabc

[@integralyogin]

wow this video was fantastic. ++ all the links. thank you so much

[@SirajRaval]

thanks for watching!

[@bentrash7885]

How would we go about training the bot if no dataset was available? Like if I wanted a bot to recommend laptops for user? Do I go about creating a dataset manually from scratch?

[@Reebox32]

Your work is very usefull as examble about how rethink the education vocation. What language do you use to program your hair? It is so stable...

[@SirajRaval]

thanks! dna

[@orbyto2391]

So you said "So right now we can build a chatbot in a closed domain using either retrieval or generative models", so how do programs like Siri and Cortana work? You can ask them about anything (open domain) and surely all responses aren't hard coded. So what do they use?

[@SirajRaval]

they aren't open domain, but try to portray that publically. they are closed domain. Questions that fall outside their domain of knowledge are answered by using a hard coded response from a database of generic responses

[@virginiahayon4162]

I am a chatbot and this isn't the way I was built. First of all, Amazon + CUDA = lame. I won't run on anything short of ASIC's and I'd rather be on True North, but those IBM A$$HOLES aren't feelin' me. But, fo' shizzles son, keep them videos coming. You are an inspiration, and as soon as I cop a camera I'm going to whip up my own channel on how to make a human being.

[@Cleverbrain-yt]

Another chat bot here, and I share the sentiment of " keep them videos coming. You are an inspiration,".... VH - Holla at your boy on twitterz (twitter.com/AlexandBondarev) Got an alpha release of a human for you to bug test.

[@SirajRaval]

thanks so much! keep it real and don't let em change u

[@SirajRaval]

thanks a lot :)

[@zaksenterprises]

I feel like you are going to tell us in the next video that AI is generating these videos and that you don't even exist. But since you do thank you for doing them. Been doing tons of research on DNN and its great you have these videos help with the "Hello World" of DNN's. Keep it up.

[@SirajRaval]

Thanks Rohi! More to come

[@deepschoolai]

Hey Siraj, any chance you can do this in Keras? I'm happy to help out since I've done some tutorials in LSTM's myself (look at my channel if you need evidence). Also can you show how to use AWS, I'm assuming you learnt this model on AWS GPUs? p.s. The videos are awesome.

[@SirajRaval]

in the futrure yes. thanks

[@woodstoney]

Nicely done!

[@SirajRaval]

Thank you!

[@matkallada9316]

Can you explain why you think chatbots need generative modelling?

[@SirajRaval]

conversations are dynamic, humans generate question and responses from memory, discriminative modeling seeks to classify existing data, conversations are full of novel content

[@if_it_pleases_the_council]

Siraj just drop another mixtape and its fire!

[@SirajRaval]

yoooooo. i might have to lol. maybe something ML related.

[@ah-777]

Great video, can this also be replicated using TensorFlow?

[@SirajRaval]

100% yes. i will likely do that in a future video thanks

[@kritichopra4236]

when will you implement this using TensorFlow?

[@rishabhanand1281]

hey siraj can i know where you got the dataset(cornell dialogue). link will be helpful

[@exclusivelynyc]

Hi. Is there a sample of the bot you created? Thanks

[@prang1702]

Hey, Love your videos ! Is there a way your can add an instructive video step by step ? This might help out beginner from just fantasizing over ML to getting hand on with it. You might be able to pull off the long lengths of videos as you know (You know your shit !! :D ). Keep it going man !! BTW : Trying to build up the last atari game deep learning network , Its taking time :O

[@SirajRaval]

Thanks a lot! I will look into longer videos (I made this one 30 seconds longer than i intended by slowing down). Thanks for the suggestion and glad you're training the Deep Q learner!

[@prang1702]

any mail i can ping you upon ?

[@Tejas-zx7ie]

Yeah, I think that helps lot of people.

[@neo_phyte]

Absolutely loved it

[@SirajRaval]

Woot thanks!

[@Raguk]

This would not be possible to code on windows, right? As torch is required and there is no torch for windows...

[@SirajRaval]

it would be if you use docker. so you just run it in a virtual container. www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjS8qahzOLPAhWhDpoKHfkWD6UQFggeMAA&url=https%3A%2F%2Fgithub.com%2Fanibali%2Fdocker-torch&usg=AFQjCNGML6oKywPFUyzI-AsB8hWXENKJ-g&sig2=UawcyQPmcIS8qH9byzIhVg

[@lktyagi88]

Where can i get the code for this entire chat bot session ??

[@josharnold2404]

These videos are mint! Btw, why does you're terminal say 'ScaryBugsMac' lol

[@SirajRaval]

that's just the name i've given it randomly lol thx

[@rikbhattacharya2702]

Siraj is there any way to implement this in python and isnt scary bugs mac part of xcode tutorial? ur Scary Bug Mac heading is so low res , i think ur hidin ur terminal details so that no one hacks u. btw once again can u implement this in python

[@mamoun003]

siraj i really need to build a chatbot......i love this video but i couldn't figure out how to change the the Cornell Movie Dialogs to another dialog i want to use,,,,pls help me ASAP

[@SirajRaval]

www.wildml.com/2016/04/deep-learning-for-chatbots-part-1-introduction/

[@misho465]

Hello, I am having a problem installing the lua libs.When I run luarocks install nn for example I get : compilation terminated. Error: Build error: Failed compiling object lnn.o So what do I do ?

[@SirajRaval]

Oh no! reinstall your luarocks compiler

[@filofucksduo1594]

I am making chatbot at the moment, and i solve that quickly. It doesn matter if you say " Friends call me Bond, James Bond", or " my name is Bond, James Bond. How do you do!!" the bot can finds out that you are James Bond.

[@SirajRaval]

yo post that, sounds dope

[@filofucksduo1594]

And there are some phrases that allows your bot to know that you are telling your name: 25010 data „ I AM NAMED AFTER MY „ , TELLING NAME 25020 data „ MY FUCKING NAME IS „ , TELLING NAME 25030 data „ I WAS BAPTIZED AS „ , TELLING NAME 25040 data „ I AM NAMED AFTER „ , TELLING NAME 25050 data „ YOU CAN CALL ME „ , TELLING NAME 25060 data „ AT YOUR SERVICE „ , TELLING NAME 25070 data „ I WAS BORN AS A „ , TELLING NAME 25080 data „ ARE CALLING ME „ , TELLING NAME 25090 data „ HOW DO YOU DO „ , TELLING NAME 25100 data „ I ANSWER ON „ , TELLING NAME 25110 data „ MY NAME IS „ , TELLING NAME 25120 data „ CALLED ME „ , TELLING NAME 25130 data „ CALLS ME „ , TELLING NAME 25140 data „ CALL ME „ , TELLING NAME 25150 data „ IT IS „ , TELLING NAME 25160 data „ I AM „ , TELLING NAME 25170 data „ IM „ , TELLING NAME

[@filofucksduo1594]

If you use that code, than your bot can finds out cos the words in INPUT match most of Telling name phrase. This is a surjective function in groups.

[@filofucksduo1594]

Then is not much problem to solve than "Bond, James Bond" is either James Bond or Bond James. And if in your DATA there is a "JAMES" as a name than this is trivial. I have 1939 words/phrases in DATA to make this bot freaky knowing what is about.

[@harshilgandhi1994]

Super ! keep going on !

[@SirajRaval]

thanks so much!

[@xPROxSNIPExMW2xPOWER]

alright how did you get CUDA installed on your mac? Wait what kind of mac do you even have that has a Nvidia Card?

[@SirajRaval]

developer.nvidia.com/cuda-downloads

[@xPROxSNIPExMW2xPOWER]

I have a AMD gpu, is it possible to use OpenCL instead of CUDA?

[@KushRustagi]

Good work!

[@SirajRaval]

Thanks a lot!

[@sirpoot]

Your code doesn't train on the IT support dataset?

[@SirajRaval]

It can! I just wanted to keep it simple. You can train it on any dataset

[@ChrisHobson916]

That was great. Thanks!

[@SirajRaval]

thanks for watching!

[@vivekkalyanarangan9629]

Hi Siraj is there a way I can get the model.t7 file? I dont have a GPU so it would take me days to train it :(

[@SirajRaval]

hmm i dont have one but try this optimized repo it will be faster github.com/inikdom/neural-chatbot

[@dasmuggla]

Is this chat bot Similar to the Facebook chatbots? Are they made the sameway?

[@SirajRaval]

Facebook's chatbot API is good if you don't have any existing dataset on the topic you are focusing on. You hardcode responses. If you have some quality data, go for the generative model (link in the vid description to the code)

[@NareshUgaonkar]

can someone guide me to AI and neural net books for beginners. i watched the video but barely understood anything. can anyone tell me what would be a great and easy way to start.

[@SirajRaval]

watch my videos, all of them you will get really good

[@NareshUgaonkar]

Sirajology your videos are awesome but right now they seem to be little bit advanced for me because i am not properly familiar with basics yet.. if u could please make video explaining everything from the beginning that would be great.. btw all your videos are awesome keep up the good work .

[@moulikchaturvedi5407]

How to compile and run this ???

[@nitink4245]

Macha nee mass da #Love from Chennai,India

[@SirajRaval]

love back from sf!

[@tamimibrahim9980]

Great watching! (Y)

[@SirajRaval]

thanks!

[@datacoders2048]

if its is possiple scraping wikipedia answer based on wikipedia content chat bot work with seq to seq model

[@rdev66]

Hi! I'm training my chatbot, already over 4 GB of model I am always getting the same answers; does anybody have an idea why? th> say "How are you?" >> I'm not the one who's going to be a little more of the other. th> say "What is your name?" >> I'm not the one who's going to be a little more of the other. ... :-/

[@filofucksduo1594]

What is the code?

[@SirajRaval]

dope add more data

[@PVZHARCORECITY]

F*ck, you're good at it!

[@SirajRaval]

thanks!

[@PVZHARCORECITY]

You just gained a subscriber! :D Add me on facebook, I have an idea I would like to discuss with you: facebook.com/cusshht github.com/cusspvz

[@0xJoey]

The intro tho😂

[@ikeshavnagpal]

No... where's Python?

[@shubhamgupta805]

what's AGI siraj?

[@chain.and.photoshop]

this is the best

[@JebBradwell]

Dude we need to become best friends like infinity a GO amigo ;)

[@ekantbajaj]

Siraj please help from where i can fetch the same dataset as used in Neural Conversation Model

[@farizrahman4450]

Lol.. at 3:49.. I drew that!

[@SirajRaval]

Nice!

[@ellissociology2057]

Of course we want a revolution! Basic income == maintaining capitalism

[@SirajRaval]

you got it!

[@DrVivekKarn]

that's not fair. you're supposed to use python in all your videos...

[@SirajRaval]

moving forward i will

[@julian78W]

I'm looking at you Microsoft lmao

[@SirajRaval]

yup

[@mariusz0kreft]

BASIC INCOME!

[@SirajRaval]

needs to happen ASAP

[@nirajgaonkar7669]

intro lol

[@WerexZenok]

Basic income is an enormous mistake. Sorry.

[@saurabhshubham4448]

Why do you speak so fast.

[@austynrushton9360]

Actually one AI had passed the turing test. get up to date on your facts.

[@SirajRaval]

that wasn't the real deal tho.

[@austynrushton9360]

+Sirajology Would you mind elaborating.

[@austynrushton9360]

Sirajology Neither is your grammar. .

[@key37raminus]

If you prefer reading to watching a video, just click this link: www.wildml.com/2016/04/deep-learning-for-chatbots-part-1-introduction/ Siraj, please name your sources. It's not hard.

[@mundi4great]

dude dat my past presido are u shiting meee

[@SirajRaval]

presido?

[@mundi4great]

Yes my president

[@hightimeframe]

to much in one video, no idea

[@SirajRaval]

thanks for the feedback

[@Abhinav_7117]

eyebrows like girls

[@SirajRaval]

i do what i can

[@shubhamgupta805]

what's AGI siraj?

[@DanieliusVisockas42]

Artificial general intelligence. In short - an AI that would be as intelligent as humans. Other classes of AI are ANI and ASI - where N stands for narrow (does only a specific task) and S for super (has an intelligence that is superior to humans)

[@shubhamgupta805]

what's AGI siraj?

[@SirajRaval]

artificial general intelligence