Easy Peasy Machine Learning Squeezy.

Project Reflection:

Wow. Looking back over this I honestly didn’t expect to do as much as I did. Not only did I use a new way of learning material faster to great success but I dived into something I only knew very surface knowledge about and found a lot of interconnected and rich knowledge on an emerging field. Making sense of that force me to reach out to others and further helped me to learn in a new way. (Normally I’m very solitary when I learn new things) What surprised me most was even though figuring out how to make sense of machine learning was frustrating, it was absolutely one of the most fascinating and engaging things I’ve worked on in a while. The concepts made me think about their implications and also fueled my long starved love of math in a unique way. This project was such a shot in the dark for me but I’m so very happy that I did it. I would loved to have been more consistent but I knew that wasn’t probably going to happen from the outset. I look forward to taking what I’ve learned and touched on and continuing to explore it from here. Perhaps I’ll work on a more tangible project or just narrow my focus and continue working on gradually as I have been. Regardless, it’s been so rewarding. \t Huge thanks to all the lovely Day Knights for their support and guidance. ♥ GG!!!

May 3rd-5th - Day 29-31: [Posted May 5th]

Notes: While I’ve had a bit on my plate, deep down I was really detached from the project after the 25th. While there is so much to learn still I feel like I need to set a new goal now as I really have done what I wanted to even though it wasn’t all the steps I was going to take to get there. Now I need to close this out and decide what my next project will be to make sure this doesn’t end here.

Videos Watched: N/A

Reading Links: N/A

April 30th and May 1st/2nd - Day 26-28: [Posted May 3rd]

Notes: Just didn’t happen these days. Either super busy or super out of it. Looking back today it’s quite disappointing as it’s not that much time and probably could have made it work. At the very least I should have thought on exactly what my next step is as I don’t want to become inactive on this now.

Videos Watched: N/A

Reading Links: N/A

April 29th - Day 25: Complete.

Notes: Finished watching the vod I had previously and thought a bit about how I want to handle this shift in the project. Still not 100% settled but there are two focuses I feel that hold my interest in machine learning currently. \t 1. I want to be able to understand current machine learning endeavors with Neural Networking within gaming. That is, I would like to understand how the current projects by OpenAI and DeepMind actually work and be able to follow along at a reasonable level. \t 2. I would like to be able to apply the knowledge myself within the gaming field and experience for myself how these things work in some manner. I think these two will allow me to focus my efforts more effectively and develop actual goals to form a better learning dynamic. Honestly, both are probably longer term goals but with them I can start forming smaller goals or project steps which will help guide my research in addition to giving me experience. \t Need to think a bit more on this but I’m also realizing writing this that I might have already reached the conclusion of the original goal for this project. Maybe it’s time to set a new project.

Videos Watched: https://www.youtube.com/watch?v=3N9phq_yZP0

Reading Links: Nothing yet.

April 28th - Day 24: N/A [Posted April 29th]

Notes: After the bad day I had previous, I decided I really needed a break from everything. I took the entire day off and made myself do new and different activities all day. My entire goal was to break up the feeling of always working so I included the DK30 as well. Though I did go for a walk and thought about my approach to ML as a whole.

April 27th - Day 23: Complete [Posted April 29th]

Notes: This wasn’t a good day for me. My sleep schedule got disrupted and I had a bad day over all. That said, I did discuss what was bothering me lately with someone knowledgeable and got some great advice. Came to realize a few things about my approach was fine up to now but needs to change at this point. I’ve had too much of a general research approach which has been good for a bit but now I really need to focus on a smaller and more manageable piece. Also, balancing off the learning with actual practice to both focus and anchor my knowledge more and keep it relevant to me will help a lot. Lots to think over but when it comes down to it, there is simply too much to learn and I would rather learn interesting things around the areas I’m interested in than a whole body of knowledge which doesn’t have much relevance to what I will do with it. \t Though I debated noting this as complete, I did take the time to reflect on things and deal with what had been bothering me. Also, I did look at and play around with the game portion a bit https://unity3d.com/machine-learning to help resolve where I go next with my machine learning. Thus, I’ll call that a win even though it’s not actual watching of vods for 30 minutes. \t Videos watched: Nothing. \t Reading Links: https://unity3d.com/machine-learning

April 26th - Day 22: Complete

Notes: I managed to do it at the end of the day. I find I’m struggling with trying to sort out the in between knowledge and getting clarity on each area with any sense of tangibility. The resources seem to either go fully into a different discipline at a high level or just simply devolve into details of how that concept is used with a whole bunch of other concepts without talking about what it actually is within machine learning. Quite frustrating feeling like both my readings and video searches are wasting a lot of time. Going to chat with some knowledgeable people to figure out how to go from here after reviewing one or two blog pieces first that show promise.

Videos watched: Watched https://www.youtube.com/watch?v=3N9phq_yZP0 posted previously.

Reading Links: Some of the wiki links but realized I needed something in between again.

April 25th - Day 21: Incomplete. [Posted April 26th]

Notes: Day was claimed by family project of building planters and then I just simply procrastinated the reset of the evening away. No bones about it, I dropped the ball there. I wasn’t that structured yesterday so that doesn’t surprise me at all.

April 24th - Day 20: Complete

Notes: Today I just read through some of the wiki. Was slow going but it’s right before bed and I’m not a fast reader. Still having some of the same problem and I think it’s just impatience now getting to me. I made so much progress and learning starting out and now that things are going deeper, it’s going to be more going over the same introductory concepts to get to the clarifying and concepts which build on this. Definitely will have to check and see if there might be other ways of approaching this better but I’m still avoiding going into the heavy details. I think to an extent this is just where I am at. I need to slow down and get a good sense of what options I have to examine further before going down one of these approaches. Overall my reading is confirming what I know, which is good, and also making clear that each area from the maps of machine learnings are different methods or ways of approaching things based on the type of problem you are trying to solve. Each one of those will have different strength, weaknesses and struggles. Those in turn dictate specific techniques, math and computational approaches. This is why each of the areas overlaps so much with each other in machine learning and also other disciplines like data mining, many forms of mathematics and statistics, and others. \t So tomorrow I’ll try for a balance of watching some of the vods and reading some of the articles with an eye to getting the specifics I don’t know out of them and see how that feels. \t Videos Watched: I didn’t actually watch any today but I found one that I forgot to post on April 21st. It was regarding the concepts of Overfitting and Underfitting. Seems one of the really good ones on this and I might go back and see other videos by this one, they have a list that seems quite technical but that may just be the naming of the vods. https://www.youtube.com/watch?v=nj_hChhSrOI

Reading Links: N/A Some of the wiki links posted last time.

April 22nd/23rd - Day 18/19: Incomplete [Posted April 24th]

Notes: These days were just not happening for me with the Easter Holiday. Frustrating but reasonable.

April 21st - Day 17: Complete. [Posted April 24th]

Notes: Today I attempted to continue the DeepLizard playlist mentioned previously but I only got the feeling like I was wasting my time due to the presentation. After searching for a bit for new videos, I realized my approach was wrong. Thinking it over I decided needed to use different resources to answer some specific questions I have regarding specific concepts which aren’t really covered in youtube videos. Therefore I looked up a bunch of wiki articles and googled for some good maps that layout the general relationships between machine learning areas. Not much actual solid reading happened but a ton of searching and gathering was done. Now that I’m going over my links I did actually find a few videos that were worth answering specific questions or general intro to a subject.

Videos Watched: This video covered reinforcement learning from a general perspective by covering what it is, is not, why it’s needed and where it fails. While the narration goes off the rails once or twice, it’s seems pretty good. Might check out the other videos they have later but this is what I was looking for at the time. https://www.youtube.com/watch?v=JgvyzIkgxF0 Was given this link by my machine learning helper. Will definitely watch it later. https://www.youtube.com/watch?v=3N9phq_yZP0

Reading Links: The following are some links which have the general machine learning maps I was looking for. (Note that I realize these aren’t that accurate but the subject itself is very fluid so there will be domains of knowledge that can be treated as separate and distinct but in practice are usually used together) https://jixta.wordpress.com/2015/07/17/machine-learning-algorithms-mindmap/ https://medium.com/datadriveninvestor/homemade-machine-learning-in-python-ed77c4d6e25b

This one seemed quite like what I’m looking for with a good balance so I’ll be reading this one first. https://vas3k.com/blog/machine_learning/ This link was linked by both OpenAI and the video I first posted above. http://karpathy.github.io/2016/05/31/rl/ One of these, for reference, were posted before and I skimmed over them and actually understood some amount of them. https://openai.com/blog/openai-five/ https://openai.com/blog/openai-baselines-ppo/ I also will be looking at the following wikis as they were the subjects from the maps I found that were by far the most important to clarifying with where my knowledge is right now. https://en.wikipedia.org/wiki/Deep_learning https://en.wikipedia.org/wiki/Reinforcement_learning https://en.wikipedia.org/wiki/Ensemble_learning https://en.wikipedia.org/wiki/Unsupervised_learning https://en.wikipedia.org/wiki/Supervised_learning https://en.wikipedia.org/wiki/Neural_network https://en.wikipedia.org/wiki/Artificial_neural_network https://en.wikipedia.org/wiki/Machine_learning https://en.wikipedia.org/wiki/Artificial_intelligence and also https://en.wikipedia.org/wiki/Feature_learning

Also, I kept this around but no clue if it’s actually going to prove useful. https://machinelearningmastery.com/parametric-and-nonparametric-machine-learning-algorithms/ and the last one here because many of the videos I’ve found seem to start with decision trees even though they really don’t seem discussed that much when you get deeper into the topic of machine learning. At least for me right now and seems rather limited. https://towardsdatascience.com/decision-tree-in-machine-learning-e380942a4c96

April 20th - Day 16: Incomplete.

Notes: Family time took priority as my sister flew in to visit. While I could do this right now, I am not as I know it will disrupt my sleep schedule if I do. Best to wind down and hit it properly tomorrow.

April 18th/19th -Day 14/15: Incomplete. [Posted April 20th]

Notes: Just didn’t do well at all these days at all. Ended up doing nothing.

April 17th - Day 13: Complete. [Posted April 20th]

Notes: Reviewed and continued discussion with another individual knowledgeable in Machine Learning. After watching a bunch of random youtube videos seeking new knowledge or approach I came to realize something really important. The actual core of machine learning and neural networks is something I already understand. So many of the concepts and terms that were just floating around were just different ways of accomplishing the same part of the “learning” process that is a neural network. To me this was huge as what seemed so massive and ever growing was actually was the same thing in a different shape. \t\tWith this realization my feeling of seeing the same videos that were different but the same now makes perfect sense. Now that I see the main conceptual layout of this subject I can work on clarifying the lines and placing concepts which do the same thing in their own box to examine in greater detail later. For example, some of the options for artificial Neural Network layers are: \t\t- Dense (or fully connected) layers - connects all inputs to all outputs \t\t- Convolution layers - more for image data \t\t- Pooling layers \t\t- Recurrent layers - more for time series data \t\t- Normalization layers, etc. These all fulfill the same part of the process but they each do so differently and can be favored depending on the data being processed.
\t I had the thought that learning neural networks would mean a slow build and many branching knowledge paths that complete changed how things worked the deeper I got into this. That however, doesn’t seem to be the case from my discussion and my experience so far. \t Therefore, I feel quite happy to start going into more depth and clarifying the relationships of concepts in my head. Already I feel way farther than I had hoped to get in 30 days to understanding Machine Learning and Artificial Neural Networks. Cannot wait to go deeper. \t Videos Watched: I quickly peeked into the following videos which were in the resources of one of the 3Blue1Brown videos from the previous series. Of the Welch Labs videos I watched the first completely and most of the second before peeking in randomly at the 3,4 14, 10, and 11. It wasn’t that I didn’t enjoy it, but it seemed more long winded version of some other approach to machine learning. Since I didn’t want another casual walk into the subject, this one was just not deep enough for my liking. https://www.youtube.com/watch?v=i8D90DkCLhI https://www.youtube.com/watch?v=2ZhQkD1QKFw https://www.youtube.com/watch?v=0cRXaORbIFA https://www.youtube.com/watch?v=sarVw-iVWgc https://www.youtube.com/watch?v=tPHImr2sFBM https://www.youtube.com/watch?v=6cvPj9dmYTo https://www.youtube.com/watch?v=biy2yU3Auc4 Then I peeked at another Welch Labs series. Still didn’t seem right so I kept searching. https://www.youtube.com/watch?v=bxe2T-V8XRs

I hit upon this series by the youtube “deeplizard” as I decided to look into unsupervised learning. https://www.youtube.com/watch?v=lEfrr0Yr684 It was enough to make me look at the series of videos below and watch the first 9 and skip the code work near the end of each video. Not super impressed with these from a presentation and explanation standpoint but they do introduce concepts and show you the code for a program which does deep learning. If you already understand the basic concepts then it does touch on some good things. Definitely a 2.0x speed watch though so far. https://www.youtube.com/playlist?list=PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU

Reading links: https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/ Backpropagation step through given to me. http://neuralnetworksanddeeplearning.com/ A resource from the 3Blue1Brown Videos https://www.mathworks.com/discovery/unsupervised-learning.html Found it interesting reference to my questions.

This were given to me as current and excellent for perspective: https://neuro.cs.ut.ee/the-use-of-embeddings-in-openai-five/ Note the first image under the “OpenAI Five network” heading. Before clicking on the image (which opens will open the pdf view without the colors.) note that the amount of image which is the actual brain. All the green is just the input layer (perception) and the motor control, blue, is the output layers. The brain is the hidden layer. Same thing I’ve already learned and this the new research. Now I need to dive into how to make each of those areas work and the options out there.

April 16th - Day 12: Complete. [Posted April 20th]

Notes: [Doing this from memory isn’t the best and I’ll be honest, I really haven’t been doing well personally for the last 5 days. Welp, on my feet again and addressing the issues now so let’s do this review as best as I can.] This day went well though I was really struggling to figure out what to do next. Even though I was strong in the math required, watching the fourth video seemed like I was violating my overall objective of staying concept level. After I reviewed the backpropagation process video I decided to just watch it anyway and see if anything clicked at the accelerated speed. It didn’t really beyond knowing that the math sounded like it mirrored the conceptual approach pretty closely. \t I decided to clarify some things by asking someone with experience in machine learning. I had a few questions mainly pertaining to the conceptual layout and the learning process of staying at a concept level. After discussing and thinking on it for a bit, I realized that was doing myself a disservice by trying to stick only to youtube videos for all the core learning. While the accelerated speed feels great and the avoidance of some of the more detailed writings is still prudent, I can see that I need the reinforcement and clarity that specific articles can provide. Moving forward I will try to work on that balance and be completely strict in avoiding more complicated resources. As long as I’m moving at a good pace and able to understand a good portion, that should be fine. \t With this in mind my plan was to watch new series and find more resources which expanded on my overall understanding. \t Videos watched: https://www.youtube.com/watch?v=Ilg3gGewQ5U&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=3 Reviewed. https://www.youtube.com/watch?v=tIeHLnjs5U8&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=4 Watched but didn’t really bother with understanding.

Also took a peek at the linear algebra series from 3Blue1Brown but felt like I still understood it well enough so I stopped. They have a multivariable calculus one which I might actually look at later though. https://www.youtube.com/watch?v=fNk_zzaMoSs&list=PLZHQObOWTQDPD3MizzM2xVFitgF8hE_ab

April 15th - Day 11: Incomplete [Posted April 16th]

Notes: Lack of sleep and low motivation claimed me this day.

April 14 - Day 10: Complete. [Posted April 16th]

Notes: Today I went over the amount I had set by almost an additional 30 minutes. Really interesting videos that I had to stop and consider a bit. Getting the layers of the concepts clear and how they all connected was my big challenge. At this point I am starting to consider how to move forward again. The next video will be the calculus and while I don’t shy away from the math, I am concerned I’m doing myself a disservice by not going into the details at all. I think I will watch the video but not worry about understanding it yet and then move on. My thought process currently is that it’s best to get the general concepts at a high level understood before going into great detail so I know what is worth the time to understand it to a given depth. I also don’t want to go too far away from machine learning yet just to understand the details of the math. If I have concerns moving forward I might just ask someone who already knows as to what I should look at next.

Videos Watched: https://www.youtube.com/watch?v=IHZwWFHWa-w&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=2 https://www.youtube.com/watch?v=Ilg3gGewQ5U&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=3

April 13 - Day 9: Complete

Notes: Today decided to start with it earlier in the day and that was definitely a great adjustment. Mind is amped up for the day and the subject matter is fascinating me in how it is expanding. The current series I’m watching by the Youtube channel 3Blue1Brown is just so amazingly good. Not only did the creator stop and give relevance to the learning in several ways but also reminded the viewer how to get the most out of the content in the process. I’m very impressed by that and just delighted to be back on this train. Additionally there is a lot of resources attached to each video. Might watch more later but also don’t want to feed into a “catch up” mentality or just plain overload myself by doing it too much. Will update if I do. Also, I will probably rewatch the video I just saw again to clarify the concepts some more.

Videos watched: https://www.youtube.com/watch?v=IHZwWFHWa-w&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&index=2

April 9-12 - Day 5-8: Incomplete.

Note: Had a crazy few days where my schedule wasn’t really my own except yesterday but I think I just needed a day off there. Not super upset as I knew this sort of thing would happen this month. Hadn’t quite expected this though so I’m not sure if there is much to learn or adjust to here. Will just keep going now and accept this loss of time here.

April 8th - Day 4: Complete

Note: Tonight was awesome even though I didn’t quite follow my plans. My day got hijacked and my afternoon and evening wasn’t quite my own. However, I still got my vod in and it was awesome. I restart the video from last time at 1.5x speed as usual and then restarted from half way in after reflecting so I could remember the perspective and information it had. This video is so awesome as it feels both accessible and deep if you take the time and know some of the topics they show as well. This series might take a few watches but I’m already feeling the pieces building on the subject.

P.S. Edit: Having a note on my scratch pads reminding me to reflect worked really well. Would have forgotten again had it not been there.

Video watched: https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi

April 7th - Day 3: Incomplete.

Notes: I left this one too long as today was very busy and exhausting in other ways. I definitely need to set a better time than late at night as I am concerned how much my brain gets amped up by the power watching before bed. Procrastination was less to do with today’s incompleteness than just exhaustion so I’m not going to beat myself up about this one. Will set immediate after dinner as the best “last chance” place but ideally I should try to fit in before then. I think maybe even breaking up into 2 or even three equal sessions across the day might be more viable on really busy days. For now though I’ll try to plan it more specifically into my day and ensure I do it right after dinner at the latest.

April 6th - Day 2: Complete.

Notes: Completed this again in the evening though I think that’s definitely not a good idea. A combination of the subject matter and the accelerated rate of what I’m listening to is making me way too alert and mentally fried before bed. That said, today went well. I did forget to reflect periodically and after each video so I’ve made a note to remind me for that. Perhaps breaking my timer up into shorter sections would be good. Definitely good that I’m doing this in 30 minute sections as the weight of the concepts is about right for taking a break after that. Today I also found that the series I was on was veering into code and away from useful topics for me. Was concerned I would end up wasting my time searching for another one but found one pretty quickly that seems awesome and a good next step. I think the understanding I gain will allow me to find the specifics I need to clarify next easier the more I watch. I’m getting better at just skipping things I already know and just going straight for those areas I need to know next. I’m not going to line up any more for that reason. We’ll see if that works or not. Videos watched: (some were quick reviews) https://www.youtube.com/watch?v=dNRT87lZYrM https://www.youtube.com/watch?v=N1mHgsZQpr4 https://www.youtube.com/watch?v=t4XLsWhj2OM https://www.youtube.com/watch?v=rSsm2_rGoug https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi {Got half way on the last link but seems good already though I’m glad I have an idea where things are headed.}

April 5th - Day 1: Complete.

Notes: Completed it this evening and covered a lot of ground in 30 minutes. Used the increased speed feature to watch everything at 1.5x and it worked very well. Also, while I’m not taking notes or creating summaries, I do think stopping at the end of short videos or periodically at ones longer than 10 minutes is definitely going to be better for retention so I’m adding that to my project. Been watching from around 2 years ago on neural network. Watched the first 6 or so. Covered a lot of the things which I knew generally about neural networks and machine learning but starting to see how they worked more specifically. Also, new terminology is being introduced. Overall, very good so far. Videos watched: https://www.youtube.com/watch?v=fvR2JySHQHo https://www.youtube.com/watch?v=ceIlLFNmViA https://www.youtube.com/watch?v=73R4cdBSKU8 https://www.youtube.com/watch?v=rSsm2_rGoug https://www.youtube.com/watch?v=4WtuN2DnLJU