After two hours of working on this I actually manage to understand the logic we need to have to reproduce the missing code. It's a pretty good exercice. Everything we need to know is in the video at every step she explains.
@@carnaciousG Sorry guys!!! I've loaded the code to my Github: github.com/MariyaSha/GradientDescent I just got an email from a nice gentleman named Larry who pointed out the link was broken! My apologies for the mess-up, I think the old version of Wayscript is no longer operational 😭😭😭 I hope it helps!! 😊 ותודה רבה עם התגובה נתי, סורי שלקח לי 3 חודש לענות 😅
I have been in ML for 6 years and that was one of the best explanations for Gradient Descent. I wish someone explained the way you explained it when I first started, amazing work!
Thank you so much Larry! I actually wish so too! it would save me almost 2 years of pondering 🤣🤣🤣 I recently reviewed my notebooks from when I studied these concepts and it was absolutely horrible! I'm not surprised people are struggling with ML so much - nobody bothers to properly explain it! it's very frustrating to look up articles and see that everybody is using the same words and giving the exact same unclear explanation... not even sure why!!! 🤔
@@PythonSimplified lol! I just looked back too, I think its the moving up and down the fish bowl analogies that made things worse🤣 and explaining Gradient Descent as some sort of magical mystical power in a black box instead of what you did and just print the process out in code :)
Maria has an elegantly simple and yet powerful style of delivery of complex matters that I’ve not seen anywhere else. It is a truly unique way of educating.. it’s a marvel
Thank you so much, Abhishek! I'm so happy to hear you liked my tutorial! 😁😁😁 I find that people are overcomplicating ML concepts for no reason at all!!! 😩 At the end of the day - everything comes down to very simple (yet sometimes a bit abstract) math formulas: new_weight[0] = only_weight[0] + l_rate*(target-prediction)*x[0] new_bias = old_bias +l_rate*(target-prediction) So why do all the articles and online academies insist on explaining this with negative derivatives, 3D graphs of a mountain-like terrain, and df/dm or df/db evaluation?? 😳 It's like everyone repeats the exact same explanation that some fancy scholar gave back in the day, while ignoring the fact that nobody truly understands what he meant!!! By the time these articles get to the chase and show you the actual formulas - you already got frustrated and left before you had a chance to even see it! (by the way, sorry about the long rant 😅 AI is not really difficult! it's just not accessible for normal people because nobody bothered to explain it simply... but then I came along! 😜) Best of luck on your AI journey!
@@PythonSimplified one more thing I want to ask how you mange to do so many things in one video you making apps in one your explaining basic loops in another video you explained ml.. OMG.... You became my role model in just few day... You are goddess of python 🤣🤣🤣
@@abhishekgaikwad6105 hahahaha thank you so much dear!! 😁 the long answer is: I'm trying to make programming and ML more accessible to regular people, who don't necessarily have the time or money to get a Computer Science degree. The short anwer is: I don't sleep much 😅 hahahaha
Your videos are more and more professional and you explain with so much detail that it's impossible not to learn. Thank you so much for sharing your knowledge with us!
Top level !!!!! Best channel, best instructor Best material, amazing editing. This channel should hit 1m before 2022. Bloody cross entropy finally smile to me after weeks of reading
hahahaha thank you Shahar! long time no seen!!! 😀 So don't even bother studding ML elsewhere - it's a recipe for disaster! hahahaha 🤣🤣🤣 It's not a difficult topic, but everybody who tries explaining it seem to have a direct interest in confusing people! or proving how good of a scientific vocabulary they have 🤪 BTW, I'm premiering a new "train a basic neural network" episode in 15 minutes, check it out here: ruclips.net/video/xpPX3fBM9dU/видео.html
I want to thank you so much as you helped me a lot from knowing nothing in machine learning to actually making an ML project for my graduation and making me realise that I want to pursue it as my future😄
Yeeeeeey! I'm so happy to hear that! 😁😁😁 Thank you so much for your fantastic comment! AI & ML are the future (and actually the present) of our society! it's an incredible field of study and I believe it's the perfect way to go when choosing a career! Best of luck on your exciting journey! 😊
WOW, I am so very glad that I started with this video rather than the materials you mentioned. It may take several replays for full understanding but I already get the idea & concept behind Gradient Descent, thanks to your awesome tutorial. I am super excited to learn more from your coming episodes.
Thank you so much! I'm super happy to hear that!! :D I was reviewing my old AI notebooks before filming this video - and I couldn't believe the amount of gibberish I wrote without really understanding its meaning! O__O And it seems that they use the exact same method of teaching anywhere else, which is even more frustrating... Anyhow! the next episode is a code-along where we see how to train a very basic model with Numpy and Pandas - it will be awesome!!! :D
Thank you! the next video is ready and good to go, I just took some extra time to work on the thumbnail 😁 It premieres tomorrow morning (Vancouver time), but since you can hardly wait - here's a link through which you can watch it right now 😉: ruclips.net/video/xpPX3fBM9dU/видео.html Enjoy! 😀
Mariya, I follow a lot of your wisdom and knowledge pills with the hope to get better at ML and AI. You're a phantastic teacher, very clear, consequential, and also beautifull to look at with all your necklaces. You cannot improve your teaching skills as you are already at the top. Though, one thing can be improved: the naming and classification of all your materials. Every time I am looking for a pill that I saw in the paste, it takes me a lot of time to find it again, even with the search abilities of youtube. That is the only issue I kindly suggest you to rethink. The rest is awsome. Thank you for your attention.
Nobody pointed mariah but you changed the target values which were in the previous videos loss function and perceptron. 0 became 1 and vice versa. very impressive presentation though. you dont expect someone so beautiful to be so intelligent :) by social laws
Pytorch is my all-times favuorite! it's very intuitive and the documentation is crystal clear! 😀 I have a few examples using it on my Github, checkout my What The Flower project: github.com/MariyaSha/FlowerImageClassifier_GUI I'm loading a pretrained state of the art network and training it on images of flowers 😉
@@PythonSimplified Yeah I've gone through that, basically I've skimmed through all your GitHub 😄 I asked if maybe you're gonna switch over. But anyways, where did you learn pytorch? I'm really keen on learning either one, but since you say pytorch and i'mma follow your series, so if suggest a resource, I'll be very thankful 😁
Thank sos much for this video. I have no words to express my gratitude. I have a question regarding the log used by the Cross Entropy Formula. Is it the logarithm base 10 or the natural logarithm, ln? When I use log base ten I get a different result than ln. Can you please help me? Thanks!
If the previous tutorial and this one it seems like there is a variation in the cross entropy loss function. I don’t whether it a mistake or a variation and I would love if you would explain it to me
@@PythonSimplified Thanks for working so hard to make us videos. I have been working on my presentation today - it is about why I like Python and covers the history of the language and converging user groups. I hope my final interview is this week.
@@Tobs_ perfect! I hope it starts with "I like Python because Mariya is teaching it" hahahahah 🤣🤣🤣 Just kidding! Best of luck on your interview!! And don't forget to mention how great Python is for AI and Data Science! 😉 this is where it really shines! Pytorch, Pandas and Numpy are simply divine!
I love the way you make this type of educational video Can you make video to show how you make this video ..i mean with which software you used. Thank you in advance
What I find so helpful is your clarity of presentation. Long shot request, could you please do a vid on K-Means clustering using python? I will sink to bribery so I will send you a bar of your favourite chocolate if you do. :)
Why am I getting a divide by zero error in the cross_entropy function with np.log10? Nevermind. I figured out I was doing "bias += bias + l_rate * (target - prediction)" instead of "bias += l_rate * (target - prediction)". Great lesson though. Thanks!
Hi Mark 😃 It depends on how your loss graph looks like. I briefly talk about it in the end of this tutorial: ruclips.net/video/xpPX3fBM9dU/видео.html So maybe start with a small number like 3 and then if your graph is descending in a very sharp angle - add more epochs! If the graph is relatively flat - you're super close! 😉
Thank you for video. I am over 60. I made program from scrap. There is some difficulty. You calculate loss two times in epoch. Individual loss I have the same. But second calculation give another average loss. Why ? Thank you.
Must run now this Monday morning, but will definitely review the recording soon. P.S. Research on Mammal neural “nets” have shown that a behavior analogous to “gradient descent” occurs in image ( vision ) and sound ( hearing ) recognition. Much of the time, I find myself stuck at a “local minimum” and must “vibrate” . . . . . .
Hahaha I'm not even surprised, David! 😁 Neural Networks are like the math behind logic and thought! Since I've started with AI, I notice more and more how alike my neural networks and I when it comes to learning! 😀 Show me visual examples - and I'll understand everything! but without examples... I'm almost always lost! As humans we start recognizing patterns at a very young age, so we must already be pre-trained when we are born! we're just like state of the art neural networks, developed by parents and God 😉
Thank you very much for your videos. I have a question though. In the video on Cross Entropy Loss you said formula was: -(y * log (w_sum) + (1-y) * log (1-w_sum)) now in this video you write the formual as -(y * log(y^) + (1 - y) * log(1 - y^)) I am a bit lost. It seems that weighted sum was replaced by prediction. can you maybe clarify how this works? Thanks!
They are one and the same! W_sum is just the weights times features (and the bias when it is not zero), and the prediction (AKA the hypothesis) is equal to W*X (and the bias). I do understand your confusion, I've seen people who also use the result of the activation function instead, but I guess it is just another manipulation done to the W_sum and if done to all is just a transformation of the entire prediction where it is easier to get what we want out of it.
I like your teaching, Do you give me some advice as I gonna to start bachelors in C.S. I need some tips and strategy as a beginner i.e. where to start 😩
Hi Kapten! :D Congrats on starting your BCS journey! If you want to prepare for it in advance - I think it would be a good idea to learn HTML + CSS and then move on to Javascript. This will help you a lot with the programming aspect :) Then there's the math... which would be quite hard to study on your own but you can give Khan Academy a try: learn about number bases, modulu, probabilities, functions, trigonometry, vectors and matrices, combinatorics and proofs. In terms of computer science concepts I would recommend focusing on data structures and data types. Good luck with school and have fun on your new adventure! :D
@@PythonSimplified Yeah I think khan academy is good idea for engineering mathematics and should I go for computer science basics, HTML and CSS with coursera paid courses ?
AMAZING VIDEO...really... can you make a simple real aplication with code using Gradient Descent? it would be great... thanks!!! My respects from Argentina
Thank you so much Matias! 😁 Your wish is about to come true in about 10 minutes, I'm premiering a new "train a basic neural network" code along using gradient descent! 😉 Check it out here: ruclips.net/video/xpPX3fBM9dU/видео.html
Thank you Udbhav 😁 You mean Snapchat?? The only social media platforms I'm on are Linkedin, Instagram and Github, I've included all my account names in the end of the video @ 12:13 😊
@@PythonSimplified well thanks Maria 😁I don't use instagram, linkedin and git hub also😂 that's why I ask because I find a way to connect with you from social media To ask some of my queries! About python programming well you are doing great work keep it up hope you got your computer science degree 😄 thanks for all your videos 👌😄 While true: Print (" Thankyou ") Have a nice day 😉
hahahahaha eto budet prosto super! 😂😂😂 Hotelos bi 4tob tak u4ili v universitetah! oni kak budto naroshno ispolzivaiut visokie slova i trudnie formuli - 4tob ne kto tolkom ne4ego ne ponyal! no iskustvinii intelekt sovsem ne trudno ponyat esli evo normalno objesnaiut! 😉 Sposibo bolshoie za prikrasnii koment Alexei! 😁
7:13 yes yes sure it go down to 0 bcoz we calculate but what it mean by airplane landing LOLcalculate once we have new values we need be this far this alt. nono we not want curve its staight line
I wished you were my teacher, I'd have understood nothing ;P - Because I wouldn't focus on the lesson hahah ... p.s: I write you comments more than I see my father, here, Instagram, everywhere ;P .. I'm not a spam ;D
9:13 i not know why im still here when dont know what these datas are and what weight is lol and where target 0 or 1 come lol. how target can change. if we have data and weight then we allready know is it 1 or 0
12:03 yes it go down. we calculate it so. if data is 1,2,3,4,5,6,7 and weight 0.1,0.2,0.2,0.7,0.3,0.7,0,4 meaning lowest lottery number last 7 week and weight is percent of how many time number has drawn. so what would be target? how predict next draw smallest number LOL yes yes data get smaller if recalculate it so. is answer who 0 first lol so what is it mean then? lol. i try understand lol. i just not get target 0 or 1 rolled with dice thing. if we allready know then we know
0:31 you have to know flying to calculate something like that lol and more like straight not curve lol its like need be in this altitude this distance with this speed from airport its straight line. landing from that point just continue have correct speed and angle that lead to runway. if everything is correct airplane land softy without doing nothing, but how you can know what is best "curve" lol and your green curve is wrong. you dont drop plane then use engines again you gradual descent "straight" not curve. yes that effective to get down engines run lower let air do it job lol
@@PythonSimplified Thank you very much; I hope you give more lessons about Neural Networks and Deep Learning/Planar data classification with one hidden layer
You look unimaginably beautiful are you artificially generated 3D graphics created by some genius for the presentation of python tutorials? if this is true, please blink 2 times in the next episode.
7:41 now we have even more data and target 0 or 1 lol still not understand why wee need recalculate data yet again i understand measure that is data then use it go do something. like landing airplane but we not need multiple only one lol we not need know what it was before we only need know what is now. we cant calculate it we measure it then do correction lol not get x(0) tables. if those are then they are why need calculate. and what it mean when calculate again! what are new data we have? new data for perioid of time in graph? to get to 0? lol xD and yes what is that weight? lol if we allready have data where weight come lol its like you measure lenght with digital caliber thats weight like 0.8 but if do it tape measure weight is 0.6? lol dont get it. dont get all datas when we only have one. if we have table then we allready have table no need recalculate lol and is it target 0 or 1 what target?
6:38 not get it. data table should allready have these. why need again calculate and all we need know is it target 0 LOL what if target we need is 1? LOL AI= automated instruction. there is no artificial intelligence. its all programmed. i just not understand data we allready have need re calculate lol and what are these datas. airplane landing? what are startting data? like predicted path? so we allready know it then no need calculate more xD
Hi Maria, first of all thank you for everything🙂
I tried to open the links to the codes and failed
I would love to receive an active link
After two hours of working on this I actually manage to understand the logic we need to have to reproduce the missing code.
It's a pretty good exercice. Everything we need to know is in the video at every step she explains.
@@carnaciousG Sorry guys!!! I've loaded the code to my Github:
github.com/MariyaSha/GradientDescent
I just got an email from a nice gentleman named Larry who pointed out the link was broken! My apologies for the mess-up, I think the old version of Wayscript is no longer operational 😭😭😭
I hope it helps!! 😊
ותודה רבה עם התגובה נתי, סורי שלקח לי 3 חודש לענות 😅
I have been in ML for 6 years and that was one of the best explanations for Gradient Descent. I wish someone explained the way you explained it when I first started, amazing work!
Thank you so much Larry! I actually wish so too! it would save me almost 2 years of pondering 🤣🤣🤣
I recently reviewed my notebooks from when I studied these concepts and it was absolutely horrible! I'm not surprised people are struggling with ML so much - nobody bothers to properly explain it! it's very frustrating to look up articles and see that everybody is using the same words and giving the exact same unclear explanation... not even sure why!!! 🤔
@@PythonSimplified lol! I just looked back too, I think its the moving up and down the fish bowl analogies that made things worse🤣 and explaining Gradient Descent as some sort of magical mystical power in a black box instead of what you did and just print the process out in code :)
Maria has an elegantly simple and yet powerful style of delivery of complex matters that I’ve not seen anywhere else. It is a truly unique way of educating.. it’s a marvel
You made it easy... I was learning this from books it took me 4 days to understand... You explained in 10 mins... I can understand your hard work 😊😊😊
Thank you so much, Abhishek! I'm so happy to hear you liked my tutorial! 😁😁😁
I find that people are overcomplicating ML concepts for no reason at all!!! 😩
At the end of the day - everything comes down to very simple (yet sometimes a bit abstract) math formulas:
new_weight[0] = only_weight[0] + l_rate*(target-prediction)*x[0]
new_bias = old_bias +l_rate*(target-prediction)
So why do all the articles and online academies insist on explaining this with negative derivatives, 3D graphs of a mountain-like terrain, and df/dm or df/db evaluation?? 😳 It's like everyone repeats the exact same explanation that some fancy scholar gave back in the day, while ignoring the fact that nobody truly understands what he meant!!!
By the time these articles get to the chase and show you the actual formulas - you already got frustrated and left before you had a chance to even see it!
(by the way, sorry about the long rant 😅 AI is not really difficult! it's just not accessible for normal people because nobody bothered to explain it simply... but then I came along! 😜)
Best of luck on your AI journey!
@@PythonSimplified one more thing I want to ask how you mange to do so many things in one video you making apps in one your explaining basic loops in another video you explained ml.. OMG.... You became my role model in just few day... You are goddess of python 🤣🤣🤣
@@abhishekgaikwad6105 hahahaha thank you so much dear!! 😁 the long answer is: I'm trying to make programming and ML more accessible to regular people, who don't necessarily have the time or money to get a Computer Science degree.
The short anwer is: I don't sleep much 😅 hahahaha
@@PythonSimplified me too its 1 o clock in morning in India and I am watching your video and thinking why I don't have brain like you 🥺🥺🤣🤣🤣
@@PythonSimplified Thank you for being HERE for US!!
You are the best teacher no teacher I have ever had :)
Your videos are more and more professional and you explain with so much detail that it's impossible not to learn. Thank you so much for sharing your knowledge with us!
Top level !!!!!
Best channel, best instructor
Best material, amazing editing.
This channel should hit 1m
before 2022.
Bloody cross entropy finally smile to me after weeks of reading
hahahaha thank you Shahar! long time no seen!!! 😀
So don't even bother studding ML elsewhere - it's a recipe for disaster! hahahaha 🤣🤣🤣
It's not a difficult topic, but everybody who tries explaining it seem to have a direct interest in confusing people! or proving how good of a scientific vocabulary they have 🤪
BTW, I'm premiering a new "train a basic neural network" episode in 15 minutes, check it out here:
ruclips.net/video/xpPX3fBM9dU/видео.html
@@PythonSimplified
Thank you, Mariya :))
I will
I have wanted to know how ML works for ages - you have got me started. Thank you.
Best explanation of GD in 12:34 mins. Perfection!
You can present complex issues in a very simple and clear way, thank you so much! Greetings from Poland :D
Thank you so much SilveRousePL! super glad you found it simple! :D
Cheers from Canada :)
I want to thank you so much as you helped me a lot from knowing nothing in machine learning to actually making an ML project for my graduation and making me realise that I want to pursue it as my future😄
Yeeeeeey! I'm so happy to hear that! 😁😁😁
Thank you so much for your fantastic comment! AI & ML are the future (and actually the present) of our society! it's an incredible field of study and I believe it's the perfect way to go when choosing a career!
Best of luck on your exciting journey! 😊
My Interest + world class education by you = success 🙌
WOW, I am so very glad that I started with this video rather than the materials you mentioned. It may take several replays for full understanding but I already get the idea & concept behind Gradient Descent, thanks to your awesome tutorial. I am super excited to learn more from your coming episodes.
Thank you so much! I'm super happy to hear that!! :D
I was reviewing my old AI notebooks before filming this video - and I couldn't believe the amount of gibberish I wrote without really understanding its meaning! O__O
And it seems that they use the exact same method of teaching anywhere else, which is even more frustrating...
Anyhow! the next episode is a code-along where we see how to train a very basic model with Numpy and Pandas - it will be awesome!!! :D
@@PythonSimplified Yay, Numpy & Pandas are my Fav couple.
this IS actually the best explanation. ive watched like 30 even from MIT videos
wow! Maria, you are a star and my favorite vlogger. Keep up the good work! thanks!
My learning rate is 100% with you :)
hahahaha thank you so much dexteuse freeman!! 😁
The explaination is beautiful as the instructor.
Best teacher ever 👍
Thank you so much Emanuele! 😁😁😁
You are freaking awesome!!! And I teach this stuff!
Thank you so much!!!! Super happy you liked it!!! 😃😃😃
Excellent Explanation with good catching examples and presentation 🤟
thank you. I can hardly wait for the next video.
Thank you! the next video is ready and good to go, I just took some extra time to work on the thumbnail 😁
It premieres tomorrow morning (Vancouver time), but since you can hardly wait - here's a link through which you can watch it right now 😉: ruclips.net/video/xpPX3fBM9dU/видео.html
Enjoy! 😀
Thanks! Never saw an explanation which was as conprehensible as yours
your explained complex thing very clear . smart and beautiful. thanks
Great! Thanks. what about the stochastic one?
explanation without derivative) it's brilliant) better than in university
Thank you, Mariya!!
You're welcome ODRACLIVE! 😁 enjoy! 😊
All those Greek letters😊thank you for omitting them and thanks for such a simple and clear explanation 🙏
thx Mariya! this class video very easy to understand
Thank you so much ShuXiaokai , glad you liked it! 😊
Mariya, I follow a lot of your wisdom and knowledge pills with the hope to get better at ML and AI. You're a phantastic teacher,
very clear, consequential, and also beautifull to look at with all your necklaces. You cannot improve your teaching skills as you are already at the top. Though, one thing can be improved: the naming and classification of all your materials. Every time I am looking for a pill that I saw in the paste, it takes me a lot of time to find it again, even with the search abilities of youtube. That is the only issue I kindly suggest you to rethink. The rest is awsome. Thank you for your attention.
Thank you so much for this series.
Thanks a lot Maria
You're absolutely welcome! 🙂
Beautiful and you know gradient decent! What a combination! The whole package! Lol
hahahaha thank you so much! 😁 wait until you hear that I'm also cooking! 🤣🤣🤣
@@PythonSimplified Cooking too!! Wow!! The Neural Network schooling Master Chef!! Lol
@@uncoded0 🤣🤣🤣🤣
@@PythonSimplified Beautiful Mariya, you just fried my brains. LOL, just joking. You have amazing talents. I am into cooking too!
Nobody pointed mariah but you changed the target values which were in the previous videos loss function and perceptron. 0 became 1 and vice versa. very impressive presentation though. you dont expect someone so beautiful to be so intelligent :) by social laws
Leaving a comment to help with the algorithm.
Thank you Mariya. You have a nice way of teaching. It reflects your depth of understanding. Where do you work? Can you share your linkedin?
Hi Mariya! What deep learning framework do you plan on using for complex models?
Pytorch is my all-times favuorite! it's very intuitive and the documentation is crystal clear! 😀
I have a few examples using it on my Github, checkout my What The Flower project:
github.com/MariyaSha/FlowerImageClassifier_GUI
I'm loading a pretrained state of the art network and training it on images of flowers 😉
@@PythonSimplified Yeah I've gone through that, basically I've skimmed through all your GitHub 😄
I asked if maybe you're gonna switch over. But anyways, where did you learn pytorch? I'm really keen on learning either one, but since you say pytorch and i'mma follow your series, so if suggest a resource, I'll be very thankful 😁
You are an absolute goddess. This video is incredible
Finally, some gradients
Thank sos much for this video. I have no words to express my gratitude. I have a question regarding the log used by the Cross Entropy Formula. Is it the logarithm base 10 or the natural logarithm, ln? When I use log base ten I get a different result than ln. Can you please help me? Thanks!
Wow, such a great video.
Thank you so much! glad you liked it! 😊
Thanku so much i am currently learning machine learning course from coursera❤️🤓
You're absolutely welcome! :D
I'm learning my BCS from Coursera (and University of London) ;)
@@PythonSimplified Can you please tell me what is your total cost for BCS i am also looking for that 😊🙏
If the previous tutorial and this one it seems like there is a variation in the cross entropy loss function. I don’t whether it a mistake or a variation and I would love if you would explain it to me
I think that in the last model the values for the Y (target) were [1, 0, 1, 0] and now Y (target) is [0, 1, 0, 1]. I guess that Mariya confussed that.
awesome explanation. thank you!
Really good thank you.
Thank you Phil, glad you liked it! 😀
40 minutes to take off, I shall make a big cup of tea now.
Thanks for joining the live chat Tobs! 😀 I hope you enjoyed your tea!
@@PythonSimplified Thanks for working so hard to make us videos. I have been working on my presentation today - it is about why I like Python and covers the history of the language and converging user groups. I hope my final interview is this week.
@@Tobs_ perfect! I hope it starts with "I like Python because Mariya is teaching it" hahahahah 🤣🤣🤣 Just kidding! Best of luck on your interview!! And don't forget to mention how great Python is for AI and Data Science! 😉 this is where it really shines! Pytorch, Pandas and Numpy are simply divine!
waw cool explanation!
Thank you so much! :D
many many thanks. perfect explanation
wow great job!
I love the way you make this type of educational video Can you make video to show how you make this video ..i mean with which software you used. Thank you in advance
if we change the step function by sigmoid function how do i know what target it predicted? how do i know if were zero or one? Greetings from Colombia
Hi from Argentina! Nice video, Mariya! I love your channel and how you explain
Excellent!! But I’m getting a broken link on the code links.
My apologies!! I've just loaded both the starter and the complete code to my Github, you can find it here 🙂:
github.com/MariyaSha/GradientDescent
What I find so helpful is your clarity of presentation. Long shot request, could you please do a vid on K-Means clustering using python? I will sink to bribery so I will send you a bar of your favourite chocolate if you do. :)
Great video Mariya. Can you explain the line on data for target ? why not np.array[0,1] but rather [0,1,0,10]. Thanks a bunch
Great content.
Nice video.
Very good video🎉💗
Thank you so much! happy to hear that! 😀
you are very underrated mam
Thank you :D!
Why am I getting a divide by zero error in the cross_entropy function with np.log10?
Nevermind. I figured out I was doing "bias += bias + l_rate * (target - prediction)" instead of "bias += l_rate * (target - prediction)".
Great lesson though. Thanks!
Thank you so much Aniruddha! :)
I'm happy to hear you worked it out and thanks for sharing your solution! :D
Thanks again. at minute 3:36, it does not seem the first feature, it seem the first data point!
Smiling face today? Nice..😊
Hahaha smiling as usual, no? 😀
@@PythonSimplified 😂
What's a good rule of thumb for determining the number of epochs you want to train over? Just acceptable average loss?
Hi Mark 😃
It depends on how your loss graph looks like. I briefly talk about it in the end of this tutorial:
ruclips.net/video/xpPX3fBM9dU/видео.html
So maybe start with a small number like 3 and then if your graph is descending in a very sharp angle - add more epochs!
If the graph is relatively flat - you're super close! 😉
@@PythonSimplified Great, thanks Mariya!
Thank you for video. I am over 60.
I made program from scrap. There is some difficulty.
You calculate loss two times in epoch. Individual loss I have the same.
But second calculation give another average loss. Why ?
Thank you.
Great!!!
Мария, спасибо вам большое за видео. Все внятно и четко объяснила, дает впечатление, что вы могли бы учить машинному обучению ребенка
hahaha spasibo bolshoie!!! 😃😃😃
mojet v budushem tak i budet!!! 😉
you here used (Stochastic Gradient Descent) is it true??
Page not found
The page you are looking for doesn't exist or has been moved. Where's the python code ??
Sorry Joep!! it is now loaded to my Github, you can find it here 🙂:
github.com/MariyaSha/GradientDescent
Must run now this Monday morning, but will definitely review the recording soon.
P.S. Research on Mammal neural “nets” have shown that a behavior analogous
to “gradient descent” occurs in image ( vision ) and sound ( hearing ) recognition.
Much of the time, I find myself stuck at a “local minimum” and must “vibrate” . . .
. . .
Hahaha I'm not even surprised, David! 😁
Neural Networks are like the math behind logic and thought! Since I've started with AI, I notice more and more how alike my neural networks and I when it comes to learning! 😀
Show me visual examples - and I'll understand everything! but without examples... I'm almost always lost!
As humans we start recognizing patterns at a very young age, so we must already be pre-trained when we are born! we're just like state of the art neural networks, developed by parents and God 😉
@@PythonSimplified I actually believe that some learning may be passed along in our DNA. Thanks for your reply !
@@davidtindell950 thank you for your comment! 😊
Thank you very much for your videos.
I have a question though.
In the video on Cross Entropy Loss you said formula was:
-(y * log (w_sum) + (1-y) * log (1-w_sum))
now in this video you write the formual as
-(y * log(y^) + (1 - y) * log(1 - y^))
I am a bit lost. It seems that weighted sum was replaced by prediction. can you maybe clarify how this works?
Thanks!
They are one and the same! W_sum is just the weights times features (and the bias when it is not zero), and the prediction (AKA the hypothesis) is equal to W*X (and the bias). I do understand your confusion, I've seen people who also use the result of the activation function instead, but I guess it is just another manipulation done to the W_sum and if done to all is just a transformation of the entire prediction where it is easier to get what we want out of it.
I like your teaching, Do you give me some advice as I gonna to start bachelors in C.S. I need some tips and strategy as a beginner i.e. where to start 😩
Hi Kapten! :D
Congrats on starting your BCS journey!
If you want to prepare for it in advance - I think it would be a good idea to learn HTML + CSS and then move on to Javascript. This will help you a lot with the programming aspect :)
Then there's the math... which would be quite hard to study on your own but you can give Khan Academy a try: learn about number bases, modulu, probabilities, functions, trigonometry, vectors and matrices, combinatorics and proofs.
In terms of computer science concepts I would recommend focusing on data structures and data types.
Good luck with school and have fun on your new adventure! :D
@@PythonSimplified Yeah I think khan academy is good idea for engineering mathematics and should I go for computer science basics, HTML and CSS with coursera paid courses ?
AMAZING VIDEO...really... can you make a simple real aplication with code using Gradient Descent? it would be great... thanks!!! My respects from Argentina
Thank you so much Matias! 😁 Your wish is about to come true in about 10 minutes, I'm premiering a new "train a basic neural network" code along using gradient descent! 😉
Check it out here:
ruclips.net/video/xpPX3fBM9dU/видео.html
@@PythonSimplified GREAT!!! I will see you today.
The phython have the cms or not madam.
how u determine bias as 0.5 value
I love you!
The way you say (hii everyone) is totally good 👍👌❤️are you on snap? And obviously video is totally great about (AI)
Thank you Udbhav 😁
You mean Snapchat?? The only social media platforms I'm on are Linkedin, Instagram and Github, I've included all my account names in the end of the video @ 12:13 😊
@@PythonSimplified well thanks Maria 😁I don't use instagram, linkedin and git hub also😂 that's why I ask because I find a way to connect with you from social media To ask some of my queries! About python programming well you are doing great work keep it up hope you got your computer science degree 😄 thanks for all your videos 👌😄
While true:
Print (" Thankyou ")
Have a nice day 😉
🙌
Ещё пара таких супер стильных видео и к Марии прибежит запыхавшийся рекрутер из Беркли, умоляя преподавать на кафедре информатики)
hahahahaha eto budet prosto super! 😂😂😂 Hotelos bi 4tob tak u4ili v universitetah! oni kak budto naroshno ispolzivaiut visokie slova i trudnie formuli - 4tob ne kto tolkom ne4ego ne ponyal! no iskustvinii intelekt sovsem ne trudno ponyat esli evo normalno objesnaiut! 😉
Sposibo bolshoie za prikrasnii koment Alexei! 😁
7:13 yes yes sure it go down to 0 bcoz we calculate but what it mean by airplane landing LOLcalculate once we have new values we need be this far this alt. nono we not want curve its staight line
its best
Thank you! 😊
I wished you were my teacher, I'd have understood nothing ;P - Because I wouldn't focus on the lesson hahah ...
p.s: I write you comments more than I see my father, here, Instagram, everywhere ;P .. I'm not a spam ;D
hahahaha thank you dear! you should see your father more often then! 😁😁😁
@@PythonSimplified hahahaha I'll do my best Mariya 🤣
i wonder if this will help me predict football scores...
Done Very Well; You Are My Favorite Programmer on The "Tube", Sweetie ~
Love The Hot Pink Nail Polish To Boot; It Is Definitely You ~
Thank you so much! 😁
I used to stay away from Barbie colours for most of my life, but now it seems I'm starting to embrace my girly side! 😅
How come you guys are grasping this, but you failed algebra and calculus? Because this is real-world, and she is awesome. That’s why.
hahahaha thank you so much for the fantastic comment Scott!! 😁😁😁
I'm super happy you liked my explanation! 😀
Refreshing that people are dressing up for RUclips videos.
I must admit that haven't seen many people who are not dressed on RUclips... maybe only in Rihanna (or other random pop star) videos 🤣🤣🤣 hahahaha
@@PythonSimplified LoLL.
Nice one
9:13 i not know why im still here when dont know what these datas are and what weight is lol and where target 0 or 1 come lol. how target can change. if we have data and weight then we allready know is it 1 or 0
🏆🏆🏆🏆🏆🏆🏆🏆❤❤❤❤❤❤❤❤❤❤❤❤
12:03 yes it go down. we calculate it so.
if data is 1,2,3,4,5,6,7 and weight 0.1,0.2,0.2,0.7,0.3,0.7,0,4
meaning lowest lottery number last 7 week and weight is percent of how many time number has drawn. so what would be target? how predict next draw smallest number LOL
yes yes data get smaller if recalculate it so. is answer who 0 first lol so what is it mean then? lol. i try understand lol. i just not get target 0 or 1 rolled with dice thing. if we allready know then we know
i love u
0:31 you have to know flying to calculate something like that lol and more like straight not curve lol its like need be in this altitude this distance with this speed from airport its straight line.
landing from that point just continue have correct speed and angle that lead to runway.
if everything is correct airplane land softy without doing nothing, but how you can know what is best "curve" lol and your green curve is wrong. you dont drop plane then use engines again you gradual descent "straight" not curve. yes that effective to get down engines run lower let air do it job lol
The code please
My apologies, the link was down!! 😭
I've now loaded everything to my Github:
github.com/MariyaSha/GradientDescent
@@PythonSimplified Thank you very much; I hope you give more lessons about Neural Networks and Deep Learning/Planar data classification with one hidden layer
You look unimaginably beautiful are you artificially generated 3D graphics created by some genius for the presentation of python tutorials? if this is true, please blink 2 times in the next episode.
hahahaha thank you GUJJJEK! 😁😁😁
I'm a real living human made in Ukraine by parents and God... or at least as far as I'm aware 😅
@@PythonSimplified so you are epic Mariya, thanks for the great tutorials
P.S. We are neighbors, I am from Poland
@@jacekk9618 nice!! I have lots of Polish friends, so I know all the important (curse) words of your language! 🤣🤣
0.68
I am done here.
It is zero point six eight and not zero point sixty eight
7:41 now we have even more data and target 0 or 1 lol still not understand why wee need recalculate data yet again i understand measure that is data then use it go do something. like landing airplane but we not need multiple only one lol we not need know what it was before we only need know what is now. we cant calculate it we measure it then do correction lol not get x(0) tables. if those are then they are why need calculate. and what it mean when calculate again! what are new data we have? new data for perioid of time in graph? to get to 0? lol xD and yes what is that weight? lol if we allready have data where weight come lol its like you measure lenght with digital caliber thats weight like 0.8 but if do it tape measure weight is 0.6? lol dont get it. dont get all datas when we only have one. if we have table then we allready have table no need recalculate lol and is it target 0 or 1 what target?
6:38 not get it. data table should allready have these. why need again calculate and all we need know is it target 0 LOL what if target we need is 1? LOL
AI= automated instruction. there is no artificial intelligence. its all programmed.
i just not understand data we allready have need re calculate lol
and what are these datas. airplane landing? what are startting data? like predicted path?
so we allready know it then no need calculate more xD
1:16 your eyes
You are beautiful like a cake, and you make everything look like a piece of cake. I am hungry now. Thank you.
А ты ничего)
Spasibo! 😀
weird doodles LOL
for me, ml means mobile legend
dang..tell me im dumb without actually telling me im dumb