How to Make an Amazing Tensorflow Chatbot Easily

[@Nahte27]

Could you train Tensorflow to create Tensorflow tutorials?!

[@SirajRaval]

yes absolutely

[@Ayelis]

How do you think he has so many? ;)

[@yotamco100]

Siraj Raval that's way too meta for me

[@ArnabAnimeshDas]

it's metaception

[@MisterPenguin42]

Siraj is actually the result of an AI tested for best KZbin host on teaching AI. He's so clever that he's passed the Turing test, but we don't know it. I, for one, welcome our new high-energy overlord!

[@salmaanakhtar6302]

How would I create my own data-set instead of using a data-set that has already been built?

[@niko2756]

this may sound like a dumb question but, after I download the data set how do I put it in the data directory?

[@SirajRaval]

there is never a dumb question. the original repo has some instructions on downloading data see this github.com/suriyadeepan/easy_seq2seq

[@niko2756]

cool thanks

[@niko2756]

hey Siraj, so i have been training for the last 48 hours and my perplexity got to 2.78 with a learning rate of 0.30 and 110000 global steps and when i test, i still just get babel, nothing coherent. I have a couple questions: does the bot remember or get better when I tested it or can that only happened during training? In other words if I chat with the bot more does it get better? Because as of right now it doesn't seem like that's the case at all. What's the best way to give the bot new interactions and retrain the model? And lastly how much longer will I have to train on the same corpus till I get a correspondence like in your video?

[@pocketman5510]

LOOL I was just thikning the exact same thing.

[@jyothishkumar3098]

me too, btw how is ur bot now?

[@athuldevin]

i am watching this video third time each of two weeks gap . siraj, you are a great man

[@mbro7126]

I love this video, I wrote out all the code but I’m confused about how to prepare the dataset and stuff can I have help please!!! Need really soon!

[@surezsivam]

Siraj, Thanks for the video, one qq, downloaded the ZIP File from Cornell dataset and it is having many txt files, but the program looking for train.enc,train.dec and test.enc files under data directory, how to make these .enc and .dec files. am i missing something here ? Please help.

[@pierrehaumahu134]

thanks for your sharing Mr.siraj. I have one question, where can i make conversations logs for my dataset? I have a lecture project to create my chatbot with the specific issue and the plan is to communicate in Indonesian. thank you

[@kerensam7082]

u can use a json file

[@beofonemind]

siraj...you rock brother...so happy to have found you. great pace ...I agree.

[@SirajRaval]

thanks Tom! Glad to have you here

[@NaturalKiller]

All very nice, but where is the download link to that blessed data set?

[@filipemtz]

Your videos are great! Really good job! The best part is that you are really agile and clear at teaching. You don't lose time, you say what must be said, and that's it. Loved the efficiency! Congratulations!

[@SirajRaval]

thanks Filipe! really appreciate it

[@khushilparmar4264]

hey Siraj bro I am just a kid and want to know how to download python 3rd generation on my windows 8 When i downloaded, program is crashing giving the reason of missing of runtime.dll please help

[@diegoantoniorosariopalomin4977]

siraj are you a data scientist yourself or do you work in tech company ? i am asking , because im starting a career in software engineering and i wanted to know what educational path would be the most fitting for someone wanting to work with neural networks

[@paulovenz8037]

i don't want him to go, his video's are the best, we should all keep watching and contributing to patreon

[@diegoantoniorosariopalomin4977]

also , your video doesnt works on linux web browsers

[@jcims]

I'm watching it right now in Chrome on Ubuntu 14.04

[@deepbits7312]

@Diego Antonio Rosario Palomino I'm pretty sure his video ain't much different from other youtube videos. Are all other videos playing for ya?

[@diegoantoniorosariopalomin4977]

Deep Bits i think that fedora doesnt uses propietary codecs , so this video may be a using a very new or old one

[@huseinzolkepli8317]

Hi siraj, on 3:02 what is the website for those chat logs? thank you. And awesome video!

[@TienTaioan]

www.cs.cornell.edu/~cristian/Cornell_Movie-Dialogs_Corpus.html

[@bithalmundra8201]

hey siraj is popular chatbots use end to end learning or they use that complex structure as you mentioned in video?

[@SirajRaval]

google assistant is end to end. siri is very likely more complex. also siri sucks so end-to-end is the way to go

[@yourdesignworld6847]

Siraj, please when you download the file what are you using after that is there a get started video to be able to keep up with your videos?

[@Flatron303]

I am really interested in this topic and i need some books/tutorials recommendations from where to start? Algorithms that are being used, how does they work, methods and all other stuffs needed for in-depth understanding?

[@jaredchong4070]

I wanna ask you something, how do you convert a string of words into floats and integer that a neural network math can read?

[@Wonderdude63]

Trained your code from github for about 12 hours, it still spits out complete jibberish. Could you make another video detailing this a bit more? Last output before I stopped it for testing was: global step 24000 learning rate 0.4950 step-time 1.41 perplexity 12.27 eval: bucket 0 perplexity 1052.47 eval: bucket 1 perplexity 1721.98 eval: bucket 2 perplexity 1224.86 eval: bucket 3 perplexity 1163.60

[@SirajRaval]

Thanks for trying it out Mark. Yeah this one takes a while to train. I'll make another video on chatbots soon that doesn't take as long to train

[@Wonderdude63]

Awesome! Can't wait :)

[@11Schollen]

awesome, can't wait! been struggling with this chatbot for ages now^^

[@gregmcclement6423]

I ran mine a bit further and still very gibberishy. One improvement to do is that "Drunk" and "drunk" map to different words. After a week I will give lower casing the words a try. global step 234600 learning rate 0.1212 step-time 1.33 perplexity 1.30 eval: bucket 0 perplexity 2493162774.88 eval: bucket 1 perplexity 1689866329.77 eval: bucket 2 perplexity 263470001.52 eval: bucket 3 perplexity 61236617.27

[@mohamedahmed-fn8qb]

can you make long video about that topic ( chatbot with end-to-end system ) because I am confused ?

[@mokshnigamsatsangi1780]

the video helped a lot, thx! but I had one doubt please, that the functions and operators are already defined in python libraries right? and its just the dataset that we provide by ourselves, right? and we have to only think where to use the functions on the right place, right?

[@aryopradiptagema6677]

Can we make an attention mechanism and feed the result to another lstm and then do a classification? and can we make it in tflearn?

[@dikduk]

Dude! You were soooooooooo early! Congrats. 🎉

[@shaarukkhaan9802]

Siraj please tell me, what are the versions of python and tensorflow that u have used in this video?

[@xXxBladeStormxXx]

@Sirajology Do you train your networks on your own laptop or do you use some cloud computing service?

[@SirajRaval]

on my laptop but this is a great question i'm going to start using more cloud computing and make videos about it

[@jcims]

Do you know if there is any advantage to running the sequences through something like syntaxnet to extract a few more features from the text or is it sufficient to just train on the raw dialogue?

[@SirajRaval]

syntaxnet would definitely improve it yeah

[@mkrishnacharan]

What is the impact of this approach on the domain specific chatbots like HR Services, Q&A

[@jornprenger7926]

hey siraj, I would love to make neural networks, but have no idea where to start. what did you begin with?

[@FrazerKirkman]

how can you split conversations in to encoder and decoder data? You defined encoder as one side of the conversation, and decoder as the responses, but aren't most conversations questions, prompts, suggestions and responses from both sides? I'm assuming this dataset is only conversations between 2 people. What would you do if you were using conversations with multiple people?

[@SirajRaval]

the chatbot would be modeled in the style of 1 person and can be applied to multiple people, this may help suriyadeepan.github.io/2016-06-28-easy-seq2seq/

[@muthumuthu9797]

Hi siraj what is the difference of dialogueflow and tensorflow for creating the chatbot

[@ElSenorEls]

Are there any better fitting models for chat bots?

[@niiosae1315]

What are the dependencies needed to make the code work? I seem to be getting errors on data_utils and seq2seq_model

[@M0481]

Is the deadline at the end of December 8th? As in December 8th 23:59 OR December 8th 00:00?

[@SirajRaval]

December 8th at 12 noon PST

[@mdyeasinarafath4450]

Great video sir! In your opinion which is better for Text Processing: NLTK or Tensorflow??

[@luis96xd]

I want to know, too :O

[@exclusivelynyc]

Hey Siraj, can this process like the replika app for iPhone? If not, how would you go about making that app? Thanks

[@Endlessthinkerz]

Hey how would I go about creating a program that has a database or data set composing of different sciences that I could use to collaborate with?

[@pranavkirtani]

Hello,I am just getting started with machine learning,which machine learning library would you recommend for a newbie?

[@lol1145]

Siraj, you are the best. I love it how you always put some jokes in your videos, makes em a joy to watch !

[@BrianGarr]

THis was a great way to start. Struggled my way through loading python, and pycharm. Using ''with tf.Session()" as I could not figure out where config file was. Getting a NameError: name 'create_model' is not defined error, and calling it a day.

[@machapuis]

Question: I am working on an interactive diary (inspired directly from Eliza) but I would like to implement Tensorflow to make it more interesting. The chatbot will learn from my own dataset and ask questions relevant to what I am writing. How can I train the model on my own input? Do I create a short corpus to get started and then feed the new input to retrain the model?

[@M0481]

Before helping you out, I want to say that I really like that idea! Super interesting :). Although I am a beginner in ML I do believe your idea is correct. It's basically like any chatbot, although I have to admit that it will probably take a long time for it to actually generate accurate questions. Maybe it's a good idea to first train it with text which has a similar style/sentiment as yours? On top of that you could also train it to ask basic questions at first so that it can form 'basic' sentences and then further train it with your own data. I hope that helped! Good luck :)

[@machapuis]

Mick thank you :)

[@niko2756]

yeah i love that as well and have been wanting to work on something similar. all of the small projects i am doing currently is all about just that.

[@niko2756]

we should correspond somehow

[@Scroll_Keeper]

I am personally working on a Discord chatbot which I would like to do something similar. If you find a way of doing this please let me know, and I shall let you know if I do the same! :)

[@timay9220]

Waoh! And to think this came five years before chatGPT. Visionary.

[@TheKulasTV]

Where can I download customer service chat logs? I would like to try that out! :D

[@satyak1337]

I am trying to build model for bills transcript. Input is pdf file -> OCR -> key value pair What network model i should be using ?Context vector or CNN?

[@countsponge0147]

what version is being uesd here ? i would like to know bc 2.0.0 wont work

[@countsponge0147]

plz help

[@GGGamingYoutube]

hi, i couldnt find the corpus file it just redirects to a python library called convokit can anybody help

[@zearthus7089]

is the new dynamic sequnce to sequence still apply to this video?

[@SatyakiSanyals]

Siraj, please tell me what the data_utils.py and seq2seq_model.py do... I'm confused...

[@LaFourmiOptimiste]

Why "hopefully Transcendance is not included in that list" when talking about the movie zip x) ?

[@rossjpg]

can a bot chat in PUBLIC posts in facebook in state of the mesenger chatbox? like, respond to other users in the comments section of each post replacing the company community manager to answer general questions or redirect the asker to the private conversation??????

[@SirajRaval]

what Manu said

[@lgsoftwares7093]

How to query data from database and make a part of the conversation, can you post some reference ?

[@SirajRaval]

depends on the type of database, if SQL try blog.hubspot.com/marketing/sql-tutorial-introduction

[@lgsoftwares7093]

Hi siraj, I know SQL ! . I need to know how to use sql or nosql to get and create responses using tensorflow on run time. Suppose, i need a lot of calculation in order to bring response based on a request. Sorry for confusion. I need some examples if you know or have. so tensorflow + nosql = logical chat responses. Thats what i need!!

[@nmstoker]

om prakash Sharma I'd welcome Siraj's thoughts but I suspect this would be challenging with this kind of deep learning as you'll need a large training set which associated the input questions to the SQL outputs (and even then you'd need a way to turn that into chatbot output if you want "human" output and not just a table). One alternative approach is to look at the sort of ML that API.Ai does, pulling out the "intent" along with keywords. You could then have a selection of queries for the intents and parameterise them based on the keywords recognised (mapping to WHERE clause or fields for the SELECT statement). This does tend to go against the approach Siraj advocated here, but I've had some basic success this way (no plans to open source it just yet, sorry!)

[@MrRandompersondude1]

I am learning about neural networks and deeplearning for the first time as of a few days ago, the question I come bearing, (after a little research), is whether or not python 3.6 is viable for tensorflow? I have heard some discussions in which it has been stated that 3.6 is not as of yet supported and that a lot of people are still using 2.7, if you (or anyone else) could let me know it would be highly appreciated.

[@br1900s]

Great work. I love the humor it def lightens up the feeling compared to other machine learning vids. Now back to training our future robot overlords.

[@SirajRaval]

thanks Ben :)

[@osomar]

Siraj, do you have any tips on running this without a GPU, or of free resources where we can run the program?

[@SirajRaval]

i will make a vid on this soon

[@blewzooka]

Hi Siraj, when I add "import data_utils" and "import seq2seq_model" my python IDLE a no module error. I have everything installed correctly. How should I go about following your tutorial? And do you know why these modules are not being found? Thanks

[@anshulchauhan8895]

Excellent work again Siraj !! Is there a way to train a chatbot with unstructured data?

[@M0481]

My piece of advice: try and find a way to classify the data first. What data are you looking for? How can you recognize the data as data you want to train your chatbot with? Then use that data to train your chatbot.

[@anshulchauhan8895]

I've very specific dataset already available with me. I saw a video where user just uploaded a wikipedia article and after training it started answering questions to insights. The video was for hutoma platform but is not available anymore.

[@SirajRaval]

thanks Ashul. i'll make a video on reading unstructured data

[@anshulchauhan8895]

Awesome. Thanks a lot for your help!!

[@sonikajha3788]

why is there a Print out French sentence corresponding to outputs code in execute.py?? The chatbot does english conversation right??

[@jaredchong4070]

Which kind of embedding did you use for this video? Glove or Word2Vec?Sorry to ask but the code is too large for me to understand and Im new with this.

[@devangichinchankar5425]

Is this an example of retrieval based or generative based model?

[@SirajRaval]

generative

[@Dylan-qk8ss]

I love you Sriaj, your videos are a lot of fun to watch.

[@SirajRaval]

thanks Dylan much love!

[@vitsujiaura6359]

No module named 'data_utils' Help please

[@KhajaMoinuddinNadaf]

run the code from this link it is complete github.com/llSourcell/tensorflow_chatbot

[@6AxisSage]

i know its late, but pretty sure thats a pytorch module which they left out of the dependency list..

[@elliotalderson9889]

I met this error when running *test*. Does anyone know this problem? File "D:\My_document\AI\Chatbot_Conversation\tensorflow_chatbot-master\data_utils.py", line 86, in initialize_vocabulary rev_vocab.extend(f.readlines()) UnicodeDecodeError: 'utf-8' codec can't decode byte 0x92 in position 1: invalid start byte

[@mdyeasinarafath4450]

I have an anaconda distribution of python. I have my tensorflow installed... but unable to import the "data_utils" and "seq2seq_model". Can't even install them via "pip"... What should I do?

[@CHANTI8947]

this ~7 min video is probably one of the most meaningful videos out there on youtube

[@SirajRaval]

thank you !

[@zainabbhatia5222]

Hello Sir, i am a new bee in Machine Learning, not getting most of the part of the code. Can you upload a video explaining the code of how it generates encoder and decoder files

[@Endlessvoidsutidos]

hey great vid trying to work out the code though getting error -can't pickle _thread.RLock objects from def seq2seq_f method any idea how to fix?

[@ebtesamh9624]

Hi, I have question what is the best neural net ca be train in small data set

[@kerensam7082]

i tried it but it gives me a nameerror: name 'create_model' not defined

[@abstractsophy]

Keren sam maybe because you missed the name a capital letter or something or maybe you didn’t download the library

[@pankajjoshi6047]

I'm trying to make a movie recommendation chat bot, not having much luck can you suggest something for that

[@maclovesgeet]

Hi Siraj, Thanks for the video. I am trying to build a chatbot for a customer service product. I don't have chat history dataset. I am expecting this chatbot to call APIS (database query) to satisfy the user questions. How can we develop a closed domain chatbot like this?

[@SirajRaval]

Thanks for watching, this service may help you api.ai/

[@maclovesgeet]

Siraj Raval I looked at the web apis. They are great. BUT My customers do not like to let the data go out of their firewall. Any other options?

[@daveb7596]

not sure if they changed the tensorflow api or what but execute file throws errors: tensorflow(dot)python(dot)ops(dot)nn has no attribute seq2seq

[@arianvc8239]

how do you put all these pictures in the right context?

[@bill312r]

what is the website at 3:01??

[@MegaGenius012]

After it get trained, when I execute the program again I will be able to have a conversation or he will train again?

[@tylerbennett9092]

Im running into errors and wondering about the data file that holds the train.enc?

[@BhavaniRavi]

I downloaded the dataset and added it to data directory but there is no train.enc or test.enc files in it Will it be generated by the program? or should I run some other script to generate those files

[@micahjohnson7926]

I'm still waiting to see if I did it right, but you have to make train.enc, train.dec, test.end, and train.dec. I'm pretty sure you just take movie_lines.txt file from the cornell zip file, write a script to take out everything but the words (I just did a split('+++$+++')[-1] on each line), and write the lines out to the new files. I believe he said in the video that the .enc file contains lines of dialog, while the .dec file contains the responses. So basically just just write every even numbered line in movie_lines to train.enc and odd numbered lines to train.dec. Then you need to split these files in half to make test.enc and test.dec. So train.enc loses half of its lines of dialog to make test.enc, and train.dec loases half of its lines of dialog to make test.dec. Hopefully I'm right, and hopefully this helps

[@BhavaniRavi]

Thanks :)

[@brianlink391]

Nice if he was more newbie friendly!

[@brianlink391]

Thanks that made a lot of sense, unlike the video. He goes way to fast with shit hard to understand. But the way I look at is he give me good ideas to start with but beyond that hes no help. SAD!. Thanks again.

[@rishithadevathoty1917]

How to customize it according to my use case?

[@laughinglama123456]

Siraj, this might be a stupid question but is this basically psuedo code? It makes so much sense but It isnt the same as the code on github

[@SirajRaval]

never a stupid a question, yes. this code in particular i had a vision of what it should look like so i did that. we can definitely wrap the code in the github repo to be less messy

[@countsponge0147]

@@SirajRaval what version of tensor flow did you use in this vid

[@oOoHaneriChanoOo]

I'm trying to use tensorFlow on Colab, but I'm not being able to import the libraries, could you tell me how to do that?

[@e.l.5402]

import tensorflow as tf

[@oOoHaneriChanoOo]

@@e.l.5402 tensorflow is working, but I cant use speech_data (for example) nor other libraries

[@mayurbankar1500]

dude please add the full video with testing it will help us how to do it there are people just started there machine learning

[@mattblackninja]

your style is clear and doesnt waste time! great. but before launching into this, I read the comments and many people said it just outputs gibberish. and u said> 'Yeah this one takes a while to train. I'll make another video on chatbots soon that doesn't take as long to train'. is this on the way? and what length of time should actually be required to train this one well? thanks.

[@SirajRaval]

thanks Matt! absolutely, new chatbot vid is coming within a month for sure

[@anonymous871rs]

Hey Siraj, Really enjoy these videos. How can you make a chatbot that ensures it always asks and gets answers to specific questions? For example a customer service bot that always asks for certain customer details or sales questions.

[@anonymous871rs]

Great, thanks a lot!

[@moody_moolga]

I thank god for introducing me to this channel. Great work Siraj!

[@SirajRaval]

Thanks Mayur!

[@Abhinav-tk1bt]

I remember around 8 years ago i came across a very well made chatbox made in dos which didnt require any internet connection

[@SirajRaval]

nice

[@syedzarreth845]

Can i use this to teach me learning differ language?

[@TheRealityCheckpoint]

Error : no module named seq2seq_model, please help

[@madhujita]

Hi. I need a Customer software support dataset. Any idea where I might get it? Thanks in advance. PS. Amaze tutorial

[@danielheintz4779]

How does one build an end to end devils advocate that takes input claims, locates fallacy, and then offers counter points?

[@danielheintz4779]

NVM how to make a chatbot #12 pretty much hits that nail on the head.

[@rahalmehdiabdelaziz8121]

Good video 👍 I wonder if it's possible to get a remote job in usa or europe and how to get it, a video on this topic will be great

[@VikasTradingCompany]

How much is cost for chatbot

[@shakeelabbasi3061]

Hi ,Siraj Raval may be this is too much easy question because i'm new How can i add data set on my windows machine it has some issues with windows when i add the directory may be i put directory path wrong way.

[@domnicmoses7396]

The code u show here is differs=ent from the code code u show in the github page pl help i got an error saying that config and create model are not defined pleas help me somebody I also dont want to just clone the code in the repository blindly Thanks in advance

[@alishibli4720]

Dear Siraj, you god of efficient videos.

[@vishalkrishna6344]

This aged well

[@prabintamang985]

from where can i get the data you used?

[@coktailtv8774]

Thank you Siraj for this video , but I would ask you some questions errors in python code . Can you help me ? how can I contact you ?

[@KhajaMoinuddinNadaf]

I may have missed it or something but i didnt see any link for the source code. can you provide me the link

[@vibhabadhe4843]

can you tell how get accuracy for this model?

[@CrankyCoder]

So over time, according to this a chatbot could be completely retrained if the new data constantly goes against the original training data correct? How would you go about creating bias's like. Naming your chatbot "alfred". How would you make sure that the chatbot knows that if someone says 50 times "your name is john" that it doesn't begin to believe it's name is john?

[@ebtesamh9624]

Siraj , question please if i want to build sq2seq what the program should i install and it is ok to use windows or Linux. give me steps please

[@SirajRaval]

www.tensorflow.org/get_started/os_setup

[@brotherphoenix9233]

so I'm not sure how the code in this video relates to the code in the github. It seems completely disconnected. Is there something I'm missing here? A video where you break down the file structure would be really helpful!

[@SirajRaval]

yea i did a more high level approach. i can connect the two more accurately in the future thanks

[@shashikant8490]

How much time its takes to train 20 thousand record or Model.

[@veeraragav9772]

Where can i learn these kinda AI in my masters? Suggest me a good univ othr than US univs..