I Passed the NLP CAPSTONE of the Udacity AI Nanodegree! | Learning Intelligence 26

What is up guys? Welcome back to another week of Learning Intelligence. This week I'm really excited. My main goal for the week is to complete the NLP capstone project, or at least have it submitted and ready for review. Last week I was almost, almost completed it, but this week I'm definitely going to be pushing through and getting it done. Otherwise today, before I get into that, I'm going to meet the founder of an AI company in my home city. So I'm really excited for that. The reason why I'm doing that, I've really been stepping up my LinkedIn game and actually one of a colleague, I used to work with said that she, she now does customer service for, for this company that I'm going to meet, and she said the founder would be interested in meeting me. So I just said, yeah, sure, I'll come and I'll come and hear what it takes to start an AI company because that's where I want to go in the future. As you may or may not know, I want to marry my love for health and fitness with AI in some way. I haven't worked out how just yet, but that's the trajectory I'm on. I'm gonna go meet this this lovely gentleman, and I'll let you know how it goes. It has been a phenomenal day. And by the way, really cool resource. I've been listening to the Super Data Science podcast. I'd highly recommend it. I was just talking to one of the guests actually on LinkedIn. Today has been all about relationships. And so the guy I was talking about on the podcast is Ben Taylor. So shout out to Ben. Thank you so much for your words of advice. The reason why I was talking to so many people is because, guys, I kind of came to a realization. I've been doing a lot of learning and now I'm looking to enter the field. I need to start building relationships in the world of data science, in the world of healthcare. It's one thing to sit in my room and do a lot of learning and making these videos and whatnot, and it's another thing to actually get on the ground and start making or working on projects and really helping people, getting skin in the game. The last few weeks I've been pounding out on LinkedIn, trying to be more active. And it's an absolute treasure trove. I can't, I can't talk. I've mentioned in the last few videos, I can't talk it up enough. Things are just starting to happen, like being more active in the community. I'm getting outreach from people wanting to meet up with me. Thank you so much to Ashley and to Mike. Mike's the founder of an AI company in Brisbane, Attract.ai, who use artificial intelligence and, and data science and whatnot to attract the best, best talent for teams. It's, it's actually really cool. And so he gave me some insight into what it's like actually running an AI startup and what what his advice is for me in terms of going forward in the future and what I should be looking at doing. Then I was reaching out to more people. I actually did a post. I'll put that here somewhere or something. Reaching out to the LinkedIn community saying, hey guys, I'm looking to find more posts in my feed of artificial intelligence, healthcare, and data science and whatnot. I'm currently following these people. Who would you recommend? And the response was phenomenal. Like the post went, it went out to, I don't know, 25,000 people or something like that. And So I started just messaging every single person who got tagged in. I started sending them a message saying, hey, your name got mentioned. I'm looking to break into this industry. Is there anyone else you would recommend for me? The response I got, I was having a conversation with, as I said, Ben, who was on the podcast, Nick, who's from Queensland, where I'm from, and we had a great conversation. And of course, there's so many more relationships forming and whatnot, because ultimately, guys, that's, I think, that's what's going to help you the most. So I mean, you can be the best data scientist and that will definitely help you out, or best AI engineer. But if you want to start getting into the industry, it's all about forming relationships with the right people and showing people what you, what you can do or what you've done. And that's, that's essentially the next stage of what I'm trying to do to, to get into the AI and healthcare world, is to start forming relationships and start meeting the right people and talking to the right people of where I want to end up. That's what I've been doing. That was all about today. As for the rest of today, I'm gonna go over some of my notes and do some reading and get into coding tomorrow. So Josh and I just finished some study at the library. What were you studying? Uh, Xcode, learning Swift. What language? Swift. Where are you studying that? On Udemy. Udemy? Yep. I love this, so much online learning. And you look at this car park, the library is absolutely packed. That's what I like to see. So we're off. He just got a new computer, so push the Mac. How you liking that? It's awesome. Better than the desktop? Yeah. Well, that means we can come to the library. Yeah, you can drive. So we need to go get a case for Josh's Mac. I was working on The NLP capstone project, I'm almost finished. And as I said, you know when you get to that last 10%, you've still got about 90% to go. That's where I think I'm up to. Check it out, my final model is training. I'm doing the extension version of the project, so that's why I haven't submitted it just yet. So I'm trying to train my final model on training and split test data. What does that mean? Well, I'm using sklearn's function train_test_split to split the, the original dataset into a training dataset and a test dataset to see if the model can still perform on unseen test data. There's no point having a really good model that works really well on training data, but then as soon as you move it on to test data, it, it doesn't perform too well. And I've run into some issues here in some of my other models here that I've implemented they're all landing at about 47%. And I think this is a similar issue to what was happening in last week's video. As you can see here, this one without the training test split finished with 71% accuracy or just over 70%. But if we do the training test split, it ends up with 47%. I'm going to submit it. I think the code is implemented correctly. There's probably something wrong that I'm not picking up at. I'm going to ask the reviewer, can they have a look at it? I've posted in the Slack channel, I've posted in the forums. It doesn't look like many people have done the extended extension version, so who knows, maybe I'm in uncharted waters here, maybe I'm doing it completely wrong. So I'm just going to submit it, try to get some feedback, see how I'm going, and then if all else fails, I can just reiterate and fix it myself. But check this out, look what just arrived while I'm waiting for my model to train. What's that little bad boy? Oh, you know what that means. I love you, Book Depository. Let's see what's inside. It's pretty big, I think I know what it is. Let's see if we can do this with hands. All these muscles in there hopeless at opening a cardboard box. Be right back. Box opening tool 101. I can never get the tape open on these things. Oh yes! You probably, you may have heard of this book before. It's called Cracking the Coding Interview by Gayle Lachman McDowell. I'm reading this back to front. 6th edition. And look at this. Look how thick that is. I've got a lot of work to do. Guys, as I've mentioned in this video, as I'm looking to move into the industry and stop doing as much learning— well, of course keep learning, but I want to start getting and building things and start doing. And if I want to get into the industry, This is definitely one of the books to look out for because it just goes through a whole bunch of problems that you may come up with, or you will, or you probably will come up with if you interview with any, any kind of startup or any kind of tech company and they want to test your programming skills. It's going to be something from this book or some kind of coding challenge related to what's in this book. So I'm going to try and work through some of these problems. That's why I bought the big whiteboard over there, which is 'cause it's got a lot of scribble on it, aka deep learning is like pizza. You can see on the top there. So that's what I'm looking forward to. That along with I need to get back into HackerRank. I kind of put that on pause for the time being 'cause I was working on Artificial Intelligence Nanodegree, which is my current focus. So this is definitely something I'm going to be looking to implement into my day-to-day routine after I finish the AI Master's Degree. It's time for yoga now. Namaste. Guys, it's almost time, almost time to submit, but I'm going to submit it with you in about 30 seconds. But I want to show you I fixed my error, or at least I think I did. We'll let the reviewer sort of go through it properly and let them know if I'm on the right track because I've never used this, this split_train_test_split function. Anyway, here we are. When I first did it, when I first implemented this function train_test_split from scikit-learn, uh, scikit-learn.model_selection.train_test_split that function. I did it separately. So I've even left myself an example. Note: train test split should be performed or preferred— that should be performed on X and Y data together, not separately. So I did the French sentences train, the French sentences test, and the English sentences train and the English sentences test as separate entities when I should have done them as the same one. So this is a really big, big line of code here, or not even a really like line. It's, it's two lines. But anyway, it worked. So if we go down to another model, you know how it was stuck on 47%? Well, not anymore. This one is a simple model and it reached 82% accuracy or 82.88 on the validation set. And the simple model, what did it get on the test set? 39%. I did that for the rest of the models and then I got down here. I added in a little bit of a, a little bit of a last thing to model performance on test. Interestingly enough, my final model, which was the most complex, didn't do very did well on the test set. Actually, the embedding model did the best, 44.98, but that's still pretty bad across the board. My thinking to this is the more complex model, it overfits the training data. That's what I'm thinking. It overfits the training data because there is not enough data here to do a good split or do a proper split. Maybe my split was wrong. I'm just thinking out loud right now, but I think the complex models overfits the training data and then doesn't do well on the test set. Time to submit. And that was a lot longer than 30 seconds, but let's see how this goes. I've never used it and now my trackpad doesn't work. Oh my goodness. Okay, here we go. 3, 2, 1. I confirm. Oh, I have to add some comments in here. One sec. All right, added in some comments. Major thing here is always say thanks if people are doing you a favor. Um, and so I've got thank you there twice. Main question was, have I implemented the train test split function correctly? And we're going to tick all these checkboxes here. And agree and submit project. Let's see how this goes. Is it going to work? This is live. Oh yes! We will notify you when your review is complete. So this should take about 24 hours in, in, uh, in previous experience. All right, round 2. I failed the first one. Well, I didn't fail, but I didn't pass it, but we have a review available. Let's check it out. What's going to happen? Internet, don't fail me now. Here we go. Oh yeah, meet specifications. Good job. You seem to have understood theoretically and in practice coding how RNNs work for this kind of task. And we have a few bonus resources on some more stuff I can learn. And regarding my comments, for this specific project, we don't have enough data to create new translations from scratch. That's what I thought. Not enough data. We need lots and lots more. So, there we go. A bunch of green ticks there, and I know the camera is shaky, so I'm sorry for that. Oh yeah! We passed the NLP Capstone Project. I'm incredibly stoked. We did it on the second try, so not the first try like I thought, but that's alright. That's alright. You fail, fail again, fail better, and then we got it done. I think it's time to dance. Alright, I'm sorry you had to see that, but I'm excited. Let's wrap up this video. What a great week it's been. Finished the NLP capstone project, translating a corpus of English text, albeit small, to a corpus of French text. And already, the machine I worked on knows more French than I've learnt in my entire life. So that's, that's why I love AI. Next week, we're on to the Voice User Interfaces concentration of the Udacity AI Nanodegree. Which I'm so excited for. I think Voice is going to be the next, next big platform. There's so many things that you can build on there and it just, it just saves time. Last week I asked you, what are you working on? And my favorite comment, trust me, there was so many more. It was very hard to pick, but thank you all so much for leaving a comment. It's so, it really inspires me that you're all working on some incredible stuff. And a lot of us, we're all working on a similar thing. We're all working on similar projects towards a similar goal. But Demetrios, I love the fact that you're working on Andrew's, or Andrew Ng's deep learning course on Coursera. I really love that. And then you're going on to fast.ai, which is exactly where I'm going after the AI Nanodegree. So I can't wait for that. And then of course your own projects, which is definitely the same, same trajectory that I'm heading on. This, this community is amazing. You guys are doing some really Incredible stuff. This week's question of the week is: What's your idea? How can you apply AI to any industry and make some value out of it? Maybe you're like me and you have a background in health and you want to apply AI to the health world. Maybe you're in farming and you want to apply AI to the farming world. Maybe you're not in either of those and you just want to— you have an idea of how you can push the field of AI forward. So leave a comment below. What's your idea in the world of AI? And I'll shout out my favourite comment in the next video. And as always, keep learning. The future looks good.