GENERATING TV SCRIPTS WITH DEEP LEARNING | 100 Days of Code 7

What's good, y'all? Day 26 of the 100 Days of Code. I didn't film yesterday because I was so wrecked from the bootcamp, like the 2 days and the 3 days and whatever. I haven't coded for 3 days, so I guess you could say it's not, not really 100 Days of Code, but that's alright. I'm gonna get back into it today. My goal for today is to finish my deep learning assignment. It's due— today's the 6th of June, 2017. It's due on the 8th, so I'm trying to submit it today and then see where I go from there. But that's my main goal for today. I'm going to edit some videos while watching the Apple keynote that just dropped, and then I'll get into doing my assignment, and I'll hit you up with my favorite things from the Apple keynote. It's the WWDC Conference 2017. So, we'll see you soon. Sup y'all, been coding RNNs all day, trying to finish off this Udacity deep learning project by this evening, but I haven't had a break yet, or a proper break. I had some food to eat, so I'm going to go work out, clear my head, and then come home and try to submit it tonight. I've only got a little bit to done. I'm probably about 70-80% of the way through. I gotta improve the training of my neural network so it fits the data better. Right now my training loss is about 2 after 100 or so epochs, like iterations. I want to get it to below 1. So we'll see how we go. I'll hit you up once I've reached that point and when I'm about to submit it. Hopefully it's tonight. If not, it'll be first thing tomorrow. So, catch you soon. Day 27. Good morning. So I've been training my neural network here. As you can see, it's going down, getting some batches going using AWS because my computer's nowhere near powerful enough to do this. And I'm looking to generate a new TV script of The Simpsons of a scene at Moe's Tavern based off the previous 27 seasons of The Simpsons. And so I've got some changelogs here. So I'm recording how I'm changing my parameters, how I'm changing my build RNN functions and whatnot. And here was the first output. So you can see Moe is like, what's the matter, homie? Lenny Leonard, yada yada ya. And so now I'm just trying to refine it. I should be done within the next hour or so and submit it and then it's It's all done. Yesterday afternoon, I know I was supposed to finish it, but I went, worked out. I was a bit tired after that, went and spoke with my friend. So now this morning, first thing, jump straight into it. I'm going to submit this and then do machine learning assignment later, uh, probably after a nap though, because this is pretty— as well as computationally intensive, it's brain intensive. But we'll see you soon. Moment of truth here. We're hitting it. Oh yeah, Project 3, Udacity Deep Learning Nanodegree Foundations— Foundation— Deep Learning Foundations Nanodegree submitted. If everything's all good, I'm officially past the halfway mark of the course, so I'm very excited. This is from not knowing anything about code, well, any Python code about 2 months ago. So I'm happy. I'm gonna take a break now. I'm gonna get something to eat and then get back into machine learning. I've got an assignment that was due on Sunday. I didn't get it 'cause I was at a bootcamp, but that's all right. I'm gonna get it today, this week, and then fix up the rest of the course. Lots of learning to come. Lots of learning to come. I'll catch you soon. Hopefully I can finish the machine learning assignment today too. We got there. 2 assignments done in 1 day so far. Programming assignment. This is probably the hardest one yet on Coursera. Neural network learning. Check it out in MATLAB. Oh, what have we got here? I don't like MATLAB. Probably because I'm not good at it. Had a lot of help from the forums. Check this out. Got all these tabs open, whatnot. But that's week 5 done. I'm still 3 days at least behind in machine learning course. I've got a whole bunch of classes to do. Uh, it's up to week 6 now, the machine learning course, and I think it's week 11 or 12 in the deep learning course. But that is 2 major pieces of work out for the day. I'm gonna spend the rest of the day, uh, maybe just going over some Python or getting started on the next series of classes rather than doing assessment. Well, it's time for another break. Time to get something to eat. Oh, actually, this is my coding music at the moment. Time by Hans Zimmer. I just have that on repeat. If you can't see that, I will literally just have that playing on repeat. I would have played it at least, I don't know, 50 times this morning, however long that is. 50 times 4 minutes, so I can't do math at the moment. 20 minutes, 200 minutes, yeah, few hours, 4 or 5, 4 or so hours. But time for a break. But check it out. Great job and congrats on finishing project 3. I don't really have any improvements for you. Hello world, it's the 28th of June, 2017, aka day 28, oh no, sorry, the 8th of June, 2017, aka day 28 of the 100 Days of Code. And here I am learning some sequence-to-sequence models through my Udacity Deep Learning Foundations Nanodegree. So what are sequence-to-sequence models? Well, I can't explain them to you yet because I've just started learning about them, but essentially with a— like, imagine a translator, Google Translate. You feed it a sequence of words in English and then it gives you a sequence of words back in another language, or any language into any other language. So Google's sequence— Google's translator is based off sequence-to-sequence models, uh, or a chatbot. Say you, for example, typing in something to ask a chatbot, the reply would come back in a sequence. And now a lot of chatbots like mine that I built, the Personal Trainer Chatbot, are built off rules. So it understands what you type in, uh, and responds a certain bit of text or block based on what you typed in, uh, because it's set up as a rule. But what's really hard to do is train a chatbot on a whole bunch of data and then have you type in something to the chatbot, like ask it, "How are you?" And then the chatbot creates its own reply rather than from a base of rules. And that may not be the best example, but just understand that rule-based is easiest to do and general AI is hardest to do, much like in real life. Like, AI sort of is very good, like for example, AlphaGo, the, the best Go-playing robot in the world, is really good at Go, but if you told it to ride a bike, it wouldn't be able to do it. So we're a far way off general AI, but for really specific tasks, robots are getting very good. If a human can do it in about 1 second, chances are a machine can do it as well. So more learning. That's my one goal for today, is finish off this little module in deep learning. I'm doing 6 pomodoros in deep learning today. And then some writing, and then I'll, I'll hit you up what I'm up to later this afternoon. Hello world, it's day 29 of the 100 Days of Code series, and today I'm going to be learning about Python and machine learning. So I've got Python, Learn Python the Hard Way book ready to go, editor, and machine learning classes I'll get to later. Last night I actually went to my first Brisbane AI meetup. So I've been using, trying to use meetup a lot more lately because what I'm learning is sort of new, even though it's been done for, I don't know, 30+ years or something like that, or longer, ever since the dawn of computing. But the way like that people think about it now is different to how they thought about it in the past. So going to the meetup, the Brisbane AI meetup, was really good to find some people who are I don't know, with the similar mindset to what I am. And that's a really important thing. If you want to do, do anything well, surround yourself with the right people. Like life, anything, anything in general, because you're an average of the 5 people you spend the most time with. So getting around the right people is fundamentally important. In other news, I just posted a post on Medium. How many more times So, what's it mean? I'll let you read it to find out, but essentially, the average person, if you're watching this, you'll have 30,000 days of your life, right? From birth to finish, 30,000 days, give or take a few thousand. So, if you think about it like that, how can you spend each day to the best of your ability? Don't waste your time here, you only get one life. And I know that might sound sad to some, but to me it's extremely motivating. So, check it out. I'll put the link in the description, medium.com/@MrMRDBourke, B-O-U-R-K-E. But for now, I'm going to get back into code. I'll catch you later this afternoon when I've, when I've done some study. What's up, y'all? I'm about to go to bed. Not much filming today, but I did do a fair bit of study, hung out with some friends this afternoon. It is my brother's birthday as well, so I didn't really want to film too much. I wanted to spend time with the family. He's 19. I can't believe it. My second youngest brother, had to count there. I only had 3 brothers, so it's not hard, the middle brother. But a whole bunch of machine learning study. I learned how to adjust machine learning algorithms if they're not training correctly, if they're overfitting, if they're underfitting, changing the parameters, changing the datasets, things like that. Tomorrow I'm going to hike a mountain, so that's going to be really cool. Maybe do some code in the evening, but that's a wrap on day 29 of the 100 Days of Code. We'll catch you tomorrow. Day 30, we'll be officially 1 month into 100 Days of Code, and I'm learning a whole bunch. It's good to get back into a roll of studying, get back on top of things, rather than sort of saying yes to too many things. But see you tomorrow. So check this out. It isn't all sunshine and rainbows when you're doing quizzes. I got 2 out of 5 on my first attempt at this machine learning quiz using a whole bunch of notes and a whole lot more notes. So maybe I didn't understand this module as well, or maybe, I don't know, it's a bit harder than the last one. But it's weird how it works out, right? You think you understand something, but you get the results back and you don't really understand it. So that's all right, I'm going to go back to the drawing board. And yeah, I'll get— I've got 2 more attempts in this 8-hour period and see if we can get 5 out of 5, because that's my goal to get 5 out of 5. But just to show you that I don't always get it right the first time. As you see, I don't know, the last quiz I did, I got 5 out of 5 in the first try. Not this time. But we'll keep trying. We'll see where we end up, and I'll, uh, I'll hit you up with if I get it within the next 8 hours, or I might just have to try again tomorrow. Free! We made it! 5 out of 5, 100%. So, moral of the story: if you fail the first time, try again. And if you don't get the result you want the second time, try again. Third time lucky. So I'm going to keep doing more study here, and then hopefully I'll get this programming assignment finished by this evening. And yeah, that'll be my coding done for the day. And I'll probably upload this video later tonight, so this will probably be the last, day 31 will be the last day that I record in this video. That'll be the end of vlog 7, I believe?