In this episode of Learning Intelligence, Daniel Bourke documents the final stretch of Term 1 of Udacity's Artificial Intelligence Nanodegree. He walks through the challenges of Project 4—a hidden Markov model based American Sign Language recognizer—highlighting an initially disastrous word error rate (≈99%) and the debugging process aided by Udacity forum mentors and Slack. After incorporating feedback, he improves the model to a 47% recognition rate, submits the project, and receives a passing review. Bourke reflects on the value of asking for help, noting his own tendency toward stubborn independence. He then shares his broader learning habits: daily Python practice via HackerRank, participation in NaNoWriMo, regular workouts, and caring for his sunflower plant. Looking ahead, he outlines a December study plan that pairs Coursera's Deep Learning Specialization with Khan Academy's Linear Algebra course, positioning these as complementary to his AI master's roadmap. The video concludes with shout‑outs to community members, contact information, and a teaser for the next episode focusing on deep learning.
Chapters
What's going on, learners? Welcome back to Learning Intelligence, episode 13. And this one's going to be a good one because I should, by the end of this video, I will be, I will be finished term 1 of the artificial intelligence Nanodegree offered by Udacity. And today was a big day, actually. I worked through probably about 40% of project 4, so I'm about 90% done, but I've run into a little problem. See this WER here? That's word error rate. And that's 90, almost 99%. My model for detecting American Sign Language only got 2 out of 178 correct. And that's across the board for all of my functions. So, I've got to work out where the errors lie, but it's getting late here, as you can probably see that my screen was going really orange because it was Flux. So, I posted some questions in the forums. Word error rate never under 98%. And I've got a snippet of my code here, but that's, it's hidden out because of forum policy. We stick to the rules around here. Not really, but that's an important point, guys. If you need help with anything, make sure you ask for it. That's something I've got to learn in my, my own journey. I need to, to need to work on my ability to, to reach out and ask for help because usually I'm quite stubborn and I like to do things on my own and just push through. But once you run into a roadblock, levering, leveraging the help of others gets incredible things done. Human beings work better when we work together. So, if you're ever in need, reach out to us for help. You know you can reach out and ask me any question anytime. Comments, email, Twitter. That's lesson of the video, or lesson of the day actually. Reach out and ask for help when you need it. But otherwise, tomorrow I'm going to get back into it. Hopefully I've got some replies on this discussion forum. Also on the Slack channel, I've posted in there. But otherwise, if not, I'll just have to do some debugging and finish off this project. What's going on guys? I'm at the library so I can't really talk that loud. But this morning I got up and I had awesome feedback from the forum mentors and from the Slack channel. And so I've managed to implement that feedback into my project so far, and we're on the last legs. So I'm going to finish off this algorithm here. This is part of myrecognizer.py. As you can see, the forum mentors are always, always really helpful. Here's my question. Got some feedback from, I think it's Letty, Letty Rodri. I'm sorry if I'm pronouncing that name wrong, but she's amazing. She's all over everything. Every single post I check out, most of the constructive feedback is from her. So shout out to Letty. And my friend's donating blood, so that's why I'm at the library and I gotta be a bit quiet. But that's my goal for the next hour or so, and I'm gonna submit this project and we'll check in when I'm done. Check it out, guys. Submitted the final project of the Udacity Artificial Intelligence Nanodegree Term 1, and check this out, my model, it ended up recognizing sign language at about 40, 47%, the best one got. Now there are ways to improve it, but my, the best model that I was able to build, or the intelligent system with the guidelines from the course, by the way, the GitHub repo that I downloaded, that's with all the projects in the Artificial Intelligence Nanodegree, is I'm not building these models from scratch, is that they have the outline outline of the framework and I have to fill in the gaps to build, to fill in the some lines of code and whatnot and build some algorithms step by step. But the majority of them, I would say, I would say almost at least 50% of each project has already been implemented, like the background and stuff. And the students, we come in and we, we implement some top-level algorithms. So all the, all the testing, all the database and whatnot has, has been done and we come in over the top and build our own custom algorithms to solve a specific problem. So as you can see here, given some instructions, so I had to do the rec— make the recognize function, which takes in two parameters here, and then we've got some details here what those parameters are. So I'm excited, the review should come back shortly. Um, they may be— I got a message before saying that Udacity may be, uh, the reviews may be a bit slower because a lot of nanodegrees are finishing up about this time of year because it is, it is November 28th. And that's when, that's when the Udacity, the last project was actually due. You can't even see that, it's not focusing. But good news is we can tick this off in the Trello board. I'll put the link in the description, by the way. So project using hidden, use hidden Markov models to recognize American Sign Language is all done. And it should be up on my GitHub by the time I post this video. So I'll put a link down in the description and you can check it out. As for what's happening now, I'm gonna go celebrate with a workout. And then I've got to wait for my review to come back, make sure I pass Term 1, and then we'll, I don't know, we might have to spend yesterday, tomorrow, sorry, planning on what we're going to do next and what I'm going to work towards next in the artificial intelligence master's degree. Drum roll, please. Check it out, guys. We did it. We met the specifications for Project 4, build a sign language recogniser. Meets specifications. Excellent work! You've passed all the sections, and we've got some resources here for the optional part 4 of the final project of term 1 of the AI Nanodegree. I'm really excited. So today was a big day, or last, last week actually has been a pretty big day working up towards this. And I think that's the first time actually that I've passed a project on, on the first submission for all 4 of the projects so far in the AI Nanodegree. So, I'm really stoked with that. I just had a shower, I'm getting ready for bed. I checked my emails before I jump in because I was really curious to see if the review had been there or not. And it was, and we found it, passed it. So, what does that mean? That I'm doing it tomorrow? That's a good question, 'cause I'm pretty sure there's a little break in between term 1 and term 2 of the AI Nanodegree. But I'll hit you up once I find out more details. I should receive some sort of email or something saying that I've completed term 1 relatively soon. I'm not sure what the process is there. Look how much my plant has grown already. I called it Planty McPlantface, by the way. I had a poll on my Instagram story, which is actually really cool that you can poll people on Instagram now. So, if you put something on a poll, you can put a little, I don't know, pop-up or a little widget that says vote for this or vote for that. And the crowd decided to name it Planty McPlantface. Or was it going to be NALP? Plant backwards with a silent T. A Planty McPlant Face one. It's growing so quick, but I think I'm gonna have to move it into my front yard soon because it is a sunflower and sunflowers grow really big and the pot it's in is not that big. So back to code, back to what I've been up to. Today, this morning, I spent about half an hour on HackerRank. Otherwise, for the rest of the day, I've been working on my writing, on my novel that I signed up for NaNoWriMo, which is a challenge. I've got a video explaining what I was going to do. My challenge was to write 50,000 words in a month And I think— I don't think I'm going to reach 50,000 words. I'm going to end up about 40,000. So, that means I averaged about 1,300 words writing a day for the month of November, when really I should have been averaging about 1,700. Something I should have mentioned in the last clip is the power of daily practice. So, I've been doing Python every day for the past 10 days or so. I was supposed to do it all month, but I really only caught up to it about 10 days or so. And just small 25-minute blocks, one Pomodoro, and my confidence in Python has gone up dramatically. And I'm only even doing the basic challenges of HackerRank. I haven't really moved on to the advanced algorithms yet, but even just small little, little pieces of practice here and there, incredibly valuable. I spent the rest of the day planning what I'm going to do next in my AI master's degree. And so I chose for the month of December, I'm going to do the deep learning specialization on Coursera, as well as the linear algebra course on Khan Academy, which is free. And why? I want to do those in parallel because deep learning has a massive linear— well, not massive, but fairly substantial linear algebra component. It has a lot of math actually involved in it. However, a lot of the pre-existing libraries such as TensorFlow and Keras and Python and NumPy and Pandas and all that have all those fundamental math principles built into the library. So you don't necessarily need to know the math inside and out unless you're trying to, to push the field forward. And in, in the case of maybe you're doing a research PhD or something, but in my case, I'm looking to be able to use the existing libraries to build something. And that's— I'm not necessarily trying to make the deep learning model and tweak it so that it's a couple of percentage points better than the best model that's in existence. I'm happy to use what's existing and build upon the shoulders of others. So we'll stand— what's— how's Isaac Newton's saying going? I would not be able to accomplish what I did if I wasn't already standing on the shoulders of giants. So that's what I plan on doing. And of course, all of this is in the Trello board, and the rest of it is in the Medium document that I created of the artificial intelligence master's degree that that I'm doing. Both those links will be in the description. And last but not least, guys, we do have a few shoutouts of the week. So these people reached out, commented on my videos, sent me an email. There's plenty more, and if I missed you, I'm so sorry. Send me an email or Twitter and you'll be in one of my future works, or I'll make sure I get you out there. But thank you to Chris, Ben, I can't pronounce this one, but Snunham01, Nexus Wars, Avinash, and Ming. Thank you guys so much for reaching out to me. I hope I did my best to— I hope you got some value out of the questions that I answered or the comments that I replied to. If you have any questions for me at all, guys, feel free to leave a comment. If you want to see a future video of any kind, just let me know. Leave a comment below or send me an email or hit me up on Twitter. My email is daniel@mrdbourke.com and my Twitter is @MrDBourke. I'll put something here. You can check it out. All the links will be in the description, by the way. That's it for this week's episode of Learning Intelligence. Next week we'll be getting into the Coursera Deep Learning Specialization, so be sure to check it in, subscribe, and we'll see you then. And as always, keep learning.
