title
Self driving car neural network in the city - Python plays GTA with Tensor Flow p.14

description
In this self-driving car with Python video, I introduce a newer, much more challenging network and task that is driving through a city. Text tutorials and sample code: https://pythonprogramming.net/game-frames-open-cv-python-plays-gta-v/ Project Github: https://github.com/sentdex/pygta5 https://twitter.com/sentdex https://www.facebook.com/pythonprogramming.net/ https://plus.google.com/+sentdex

detail
{'title': 'Self driving car neural network in the city - Python plays GTA with Tensor Flow p.14', 'heatmap': [{'end': 702.336, 'start': 679.738, 'weight': 1}], 'summary': 'Covers training a self-driving vehicle ai using python for 13 epochs with 30 files, encountering challenges with tensorflow 1.0 and a burst water line, troubleshooting a failed graphics card, analyzing an autonomous vehicle model, and evaluating an ai model for reinforcement learning and successful self-driving car algorithms.', 'chapters': [{'end': 446.721, 'segs': [{'end': 53.531, 'src': 'embed', 'start': 24.01, 'weight': 0, 'content': [{'end': 28.613, 'text': "I ran it, although this says I ran it for 390 epochs, that's not really true.", 'start': 24.01, 'duration': 4.603}, {'end': 31.755, 'text': 'The data is actually separated into many files.', 'start': 29.433, 'duration': 2.322}, {'end': 34.856, 'text': 'So the entire data set is 30 files.', 'start': 32.375, 'duration': 2.481}, {'end': 40.5, 'text': 'So I actually ran it for 13 true epochs through the data.', 'start': 35.097, 'duration': 5.403}, {'end': 44.883, 'text': 'Yeah So these are the results.', 'start': 41.761, 'duration': 3.122}, {'end': 50.568, 'text': 'Now someone was saying that the huge fluctuations, if I recall, is due to the learning rate.', 'start': 45.043, 'duration': 5.525}, {'end': 53.531, 'text': "I'm not really ready to buy that one yet.", 'start': 51.289, 'duration': 2.242}], 'summary': 'Ran data for 13 true epochs with 30 files, skeptical about learning rate impact.', 'duration': 29.521, 'max_score': 24.01, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg24010.jpg'}, {'end': 132.814, 'src': 'embed', 'start': 76.877, 'weight': 1, 'content': [{'end': 81.22, 'text': "That does make sense to me, but that that's a solution.", 'start': 76.877, 'duration': 4.343}, {'end': 86.925, 'text': "I only think that's probably a solution once it really trains itself and gets to a higher accuracy.", 'start': 81.22, 'duration': 5.705}, {'end': 95.712, 'text': "I don't wanna lock in my results at this point, right? So I'm not really sure you definitely wanna decrease it over time.", 'start': 86.965, 'duration': 8.747}, {'end': 103.014, 'text': "But the other thing you can do is once you're kind of close to a sort of zone that you're happy with, sure, then you could decrease the learning rate.", 'start': 96.189, 'duration': 6.825}, {'end': 103.394, 'text': "I don't know.", 'start': 103.114, 'duration': 0.28}, {'end': 109.759, 'text': 'Anyway, these initial ones here are just kind of tests just to see.', 'start': 104.415, 'duration': 5.344}, {'end': 112.881, 'text': 'This is probably just a fluke because I probably just stopped it, I think.', 'start': 109.859, 'duration': 3.022}, {'end': 118.36, 'text': "But it took me a long time actually just to get something that would even train on the data that I've passed.", 'start': 113.476, 'duration': 4.884}, {'end': 125.887, 'text': "I've got basically nine input options and it's a much more unique data set than what I had before.", 'start': 118.401, 'duration': 7.486}, {'end': 128.229, 'text': 'Also, it was taking a much longer time to build it.', 'start': 126.308, 'duration': 1.921}, {'end': 132.814, 'text': "Anyway, let me start Grand Theft Auto and let's see how we did.", 'start': 129.009, 'duration': 3.805}], 'summary': 'Struggling with training accuracy, considering decreasing learning rate, unique data set with nine input options.', 'duration': 55.937, 'max_score': 76.877, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg76877.jpg'}, {'end': 247.807, 'src': 'embed', 'start': 220.867, 'weight': 3, 'content': [{'end': 224.472, 'text': 'could instead go through like the actual city of Grand Theft Auto.', 'start': 220.867, 'duration': 3.605}, {'end': 227.636, 'text': "Now it's, it's probably still going to get itself, itself stuck.", 'start': 224.512, 'duration': 3.124}, {'end': 232.281, 'text': "Also, there's so many more variables once you like leave the freeway and come into the city.", 'start': 228.137, 'duration': 4.144}, {'end': 238.823, 'text': "So variables like pedestrians, walls rather than barriers, like so on the freeway, there's the barriers.", 'start': 232.661, 'duration': 6.162}, {'end': 240.864, 'text': "And if you hit the barrier, it's like, big deal.", 'start': 238.843, 'duration': 2.021}, {'end': 243.085, 'text': 'It just, it basically keeps you on the road.', 'start': 240.924, 'duration': 2.161}, {'end': 247.807, 'text': "Whereas in the city, there's much more like walls for you to run into and stuff.", 'start': 243.965, 'duration': 3.842}], 'summary': 'Challenges increase in city driving with more variables and obstacles.', 'duration': 26.94, 'max_score': 220.867, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg220867.jpg'}], 'start': 2.162, 'title': 'Training self-driving vehicle ai in grand theft auto', 'summary': 'Covers the process of training a self-driving vehicle ai using python, running it for 13 true epochs through a data set of 30 files, and exploring the impact of learning rate on model performance. it also discusses the challenges in training a model for grand theft auto with a unique data set of nine input options and 112k training samples, encountering issues during the initial test run.', 'chapters': [{'end': 112.881, 'start': 2.162, 'title': 'Self-driving vehicles in grand theft auto', 'summary': 'Discusses the process of training a self-driving vehicle ai using python, running it for 13 true epochs through the entire data set comprising 30 files, and exploring the impact of learning rate on model performance.', 'duration': 110.719, 'highlights': ['The AI was trained for 13 true epochs through the entire 30-file data set, despite the display indicating 390 epochs.', 'The chapter explores the impact of learning rate on model performance, considering the suggestion to decrease the learning rate over time as the AI trains and approaches higher accuracy.', 'The initial training runs are considered as tests, with the intention to further refine the model in the future.']}, {'end': 446.721, 'start': 113.476, 'title': 'Training model for grand theft auto', 'summary': 'Discusses training a model for grand theft auto, using a unique data set with nine input options and 112k training samples, but with limited key presses available, leading to challenges in navigating the city environment and encountering issues during the initial test run.', 'duration': 333.245, 'highlights': ['The model uses a unique data set with nine input options and 112K training samples, but only a little over 10,000 for each key press, leading to limited key presses available for navigating the city environment.', 'Encountering issues during the initial test run, as the model struggles to navigate the city environment and crashes, indicating the challenges of exploring the city and dealing with variables like pedestrians and walls.', 'Training the model for Grand Theft Auto, doubling the alex net and changing a few parameters, but facing difficulties during the initial test run, possibly due to older software or compatibility issues with cootie and then.']}], 'duration': 444.559, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg2162.jpg', 'highlights': ['The AI was trained for 13 true epochs through the entire 30-file data set, despite the display indicating 390 epochs.', 'The model uses a unique data set with nine input options and 112K training samples, but only a little over 10,000 for each key press, leading to limited key presses available for navigating the city environment.', 'The chapter explores the impact of learning rate on model performance, considering the suggestion to decrease the learning rate over time as the AI trains and approaches higher accuracy.', 'Encountering issues during the initial test run, as the model struggles to navigate the city environment and crashes, indicating the challenges of exploring the city and dealing with variables like pedestrians and walls.', 'The initial training runs are considered as tests, with the intention to further refine the model in the future.', 'Training the model for Grand Theft Auto, doubling the alex net and changing a few parameters, but facing difficulties during the initial test run, possibly due to older software or compatibility issues with cootie and then.']}, {'end': 681.539, 'segs': [{'end': 503.163, 'src': 'embed', 'start': 446.721, 'weight': 0, 'content': [{'end': 452.327, 'text': "I mean I probably should just update everything and go through, but there's like so many little parts I just don't feel like doing it.", 'start': 446.721, 'duration': 5.606}, {'end': 462.887, 'text': "I'm up to tensorflow 1.0 and everything else that's required Other than that I'm not Oh, you're just doing me wrong, man.", 'start': 452.547, 'duration': 10.34}, {'end': 463.828, 'text': "It's like Christmas morning.", 'start': 462.907, 'duration': 0.921}, {'end': 467.11, 'text': "You're like opening the box and then there's a box inside of the box.", 'start': 463.848, 'duration': 3.262}, {'end': 474.354, 'text': 'God Work! You are killing me.', 'start': 468.551, 'duration': 5.803}, {'end': 478.417, 'text': 'I wonder if I should close out this window too or something.', 'start': 474.374, 'duration': 4.043}, {'end': 480.098, 'text': "I'll just do it.", 'start': 479.537, 'duration': 0.561}, {'end': 482.051, 'text': "You're killing me.", 'start': 481.391, 'duration': 0.66}, {'end': 485.213, 'text': 'I just want to see it run.', 'start': 482.772, 'duration': 2.441}, {'end': 489.515, 'text': "The worst part is I don't think I had enough data for this challenge.", 'start': 485.573, 'duration': 3.942}, {'end': 491.877, 'text': 'This is just like step one.', 'start': 489.976, 'duration': 1.901}, {'end': 496.639, 'text': "Like find out if it's even remotely possible to do what I wanted to do the way I wanted to do it.", 'start': 492.097, 'duration': 4.542}, {'end': 503.163, 'text': 'I just want to see it take more than like one action or two actions.', 'start': 496.659, 'duration': 6.504}], 'summary': 'Struggling with updating to tensorflow 1.0, feeling challenged by lack of data for the task.', 'duration': 56.442, 'max_score': 446.721, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg446721.jpg'}, {'end': 624.466, 'src': 'embed', 'start': 543.58, 'weight': 3, 'content': [{'end': 549.261, 'text': "For some reason this has been acting really finicky to open the command prompt in the window that you're in.", 'start': 543.58, 'duration': 5.681}, {'end': 551.742, 'text': 'It just takes multiple tries.', 'start': 549.281, 'duration': 2.461}, {'end': 556.503, 'text': 'Give me what I want.', 'start': 555.502, 'duration': 1.001}, {'end': 557.443, 'text': 'Thank you.', 'start': 556.843, 'duration': 0.6}, {'end': 560.706, 'text': 'Where did you open? Here we are.', 'start': 559.285, 'duration': 1.421}, {'end': 565.449, 'text': 'Okay Ccon slash Python 35.', 'start': 562.007, 'duration': 3.442}, {'end': 570.432, 'text': 'Python step four.', 'start': 565.449, 'duration': 4.983}, {'end': 575.856, 'text': 'All right.', 'start': 575.476, 'duration': 0.38}, {'end': 579.578, 'text': 'Moment of truth.', 'start': 578.478, 'duration': 1.1}, {'end': 581.58, 'text': "Let's do it.", 'start': 579.599, 'duration': 1.981}, {'end': 584.788, 'text': 'Please work this time.', 'start': 583.988, 'duration': 0.8}, {'end': 586.168, 'text': "I ain't got time for this.", 'start': 585.088, 'duration': 1.08}, {'end': 591.949, 'text': 'Oh, my dogs are going to probably start barking at a bunch of people in front of my house.', 'start': 589.109, 'duration': 2.84}, {'end': 594.01, 'text': 'The show must go on.', 'start': 593.09, 'duration': 0.92}, {'end': 598.771, 'text': 'Please move forward.', 'start': 597.831, 'duration': 0.94}, {'end': 601.351, 'text': "Ah, it's crashed on me again.", 'start': 599.771, 'duration': 1.58}, {'end': 604.072, 'text': 'Oh, and Python has stopped working.', 'start': 601.371, 'duration': 2.701}, {'end': 605.012, 'text': 'My goodness.', 'start': 604.352, 'duration': 0.66}, {'end': 607.972, 'text': 'Can I please? Can we please? Ah.', 'start': 605.632, 'duration': 2.34}, {'end': 618.543, 'text': 'Could not create CooDNN handle, CooDNN not initialize, meh, meh, meh, meh, meh.', 'start': 614.502, 'duration': 4.041}, {'end': 621.424, 'text': 'Could not open driver path.', 'start': 619.784, 'duration': 1.64}, {'end': 624.466, 'text': 'Could not destroy, check fail.', 'start': 622.265, 'duration': 2.201}], 'summary': 'Struggling to open command prompt for python 3.5, facing multiple crashes and errors.', 'duration': 80.886, 'max_score': 543.58, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg543580.jpg'}], 'start': 446.721, 'title': 'Challenges with tensorflow 1.0 and water line burst', 'summary': 'Delves into the frustration and challenges encountered while running tensorflow 1.0, including multiple failed attempts and technical errors. additionally, it discusses the inconvenience caused by a burst water line due to lack of communication from city authorities, resulting in a disrupted water supply.', 'chapters': [{'end': 624.466, 'start': 446.721, 'title': 'Troubleshooting tensorflow 1.0 update', 'summary': 'Highlights the frustration and challenges faced while attempting to run tensorflow 1.0, including multiple failed attempts and technical errors, expressing the desire to see the program run smoothly.', 'duration': 177.745, 'highlights': ['Encountering technical issues and frustration while attempting to run TensorFlow 1.0, experiencing multiple crashes and errors, indicating the challenges faced during the process.', 'Expressing the desire to see the program run smoothly and the frustration of encountering obstacles and errors, reflecting the determination to overcome the challenges.', 'Mentioning the need for more data for the challenge and the goal to assess the feasibility of the desired approach, indicating the initial phase of the project and the desire for progression.', 'Expressing disappointment and annoyance at the repeated crashes and technical errors, emphasizing the impact of the situation and the determination to resolve the issues.', 'Mentioning the intention to run the program in the command prompt and the frustration with the finicky behavior of the system, illustrating the persistence in attempting different methods to achieve success.']}, {'end': 681.539, 'start': 625.426, 'title': 'Water line burst and lack of communication', 'summary': 'Discusses the inconvenience caused by a burst water line, resulting in a lack of water supply due to inadequacies in communication from the city authorities.', 'duration': 56.113, 'highlights': ['The burst water line caused an intermittent lack of water supply in the neighborhood.', "The city's lack of communication regarding the water supply issue led to inconvenience for the residents.", "The fire department's rapid shut-off of water caused pipes to burst, leading to further complications."]}], 'duration': 234.818, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg446721.jpg', 'highlights': ['Experiencing frustration and technical issues while attempting to run TensorFlow 1.0, facing multiple crashes and errors.', 'Expressing determination to overcome challenges and desire to see the program run smoothly.', 'Mentioning the need for more data and the goal to assess the feasibility of the desired approach.', 'Expressing disappointment and annoyance at the repeated crashes and technical errors, emphasizing the impact of the situation.', 'Illustrating persistence in attempting different methods to achieve success, including running the program in the command prompt.']}, {'end': 860.271, 'segs': [{'end': 817.839, 'src': 'embed', 'start': 792.324, 'weight': 0, 'content': [{'end': 798.167, 'text': "i'm like trying to record and i'm running this massive model at the same time, so it's kind of problematic anyway.", 'start': 792.324, 'duration': 5.843}, {'end': 809.414, 'text': 'um so one thing i thought about, because the other one of the other models i trained, the initial one it wanted to reverse like 90 of the time.', 'start': 798.167, 'duration': 11.247}, {'end': 810.134, 'text': 'that was clearly like.', 'start': 809.414, 'duration': 0.72}, {'end': 813.956, 'text': 'the first move it learned was like oh, this is where we want to reverse.', 'start': 810.134, 'duration': 3.822}, {'end': 817.839, 'text': 'and then basically the only move it would ever do was reverse.', 'start': 813.956, 'duration': 3.883}], 'summary': 'Running a massive model while recording, facing issues. one model had a 90% reverse rate.', 'duration': 25.515, 'max_score': 792.324, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg792324.jpg'}, {'end': 866.813, 'src': 'embed', 'start': 837.398, 'weight': 1, 'content': [{'end': 842.122, 'text': "So, if there's ever a time where I'm not doing anything, where I'm reversing for more than five seconds,", 'start': 837.398, 'duration': 4.724}, {'end': 846.325, 'text': 'choose the second most common key or something like that.', 'start': 842.122, 'duration': 4.203}, {'end': 855.77, 'text': 'It also might be the case that, just weighted wise, we might need to apply a a last wait on our own to the, the data, the output layer.', 'start': 847.346, 'duration': 8.424}, {'end': 860.271, 'text': 'in theory, the neural network should have figured that one out on its own, but maybe not.', 'start': 855.77, 'duration': 4.501}, {'end': 866.813, 'text': "so. uh, those are two things that we have to look forward to, but i'm just gonna let this run a little bit longer, just because, um,", 'start': 860.271, 'duration': 6.542}], 'summary': 'Neural network may need a last weight adjustment if idle for >5 seconds.', 'duration': 29.415, 'max_score': 837.398, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg837398.jpg'}], 'start': 681.539, 'title': 'Training an ai model for game control', 'summary': "Discusses a user's training of a model to control a game character, addressing issues of excessive reversing and inactivity, and considering memory constraints and model performance.", 'chapters': [{'end': 860.271, 'start': 681.539, 'title': 'Training an ai model for game control', 'summary': "Chronicles a user's training of a model to control a game character, addressing issues of excessive reversing and inactivity, and considering memory constraints and model performance.", 'duration': 178.732, 'highlights': ['The user encounters issues with the model reversing excessively, with the initial move being to reverse 90% of the time, leading to the need for intervention.', 'The user contemplates a time limit on reversing and inactivity, proposing a solution to choose the second most common key if no action is taken for longer than five seconds.', "The user discusses the potential need to manually adjust the weighting of the output layer of the neural network to address potential limitations in the model's learning ability."]}], 'duration': 178.732, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg681539.jpg', 'highlights': ['The user encounters issues with the model reversing excessively, with the initial move being to reverse 90% of the time, leading to the need for intervention.', 'The user contemplates a time limit on reversing and inactivity, proposing a solution to choose the second most common key if no action is taken for longer than five seconds.', "The user discusses the potential need to manually adjust the weighting of the output layer of the neural network to address potential limitations in the model's learning ability."]}, {'end': 1134.94, 'segs': [{'end': 935.263, 'src': 'embed', 'start': 911.776, 'weight': 0, 'content': [{'end': 921.199, 'text': "so my graphics card just freaking dumped and uh yeah, so i i didn't really account like i tried to.", 'start': 911.776, 'duration': 9.423}, {'end': 924.799, 'text': "i made the model big enough and i was like, oh yeah, i've got plenty of vram for that,", 'start': 921.199, 'duration': 3.6}, {'end': 929.24, 'text': "but i didn't really account for a grand theft auto or recording.", 'start': 924.799, 'duration': 4.441}, {'end': 933.923, 'text': "so whoops, Anyway, let's try again.", 'start': 929.24, 'duration': 4.683}, {'end': 935.263, 'text': 'It clearly worked for a little bit.', 'start': 933.963, 'duration': 1.3}], 'summary': 'Graphics card failed during usage, underestimation of vram impact. attempting again.', 'duration': 23.487, 'max_score': 911.776, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg911776.jpg'}, {'end': 1052.789, 'src': 'embed', 'start': 977.764, 'weight': 3, 'content': [{'end': 982.588, 'text': 'What are you doing, mate? Go! Clearly no keys too often.', 'start': 977.764, 'duration': 4.824}, {'end': 983.108, 'text': "Let's go.", 'start': 982.768, 'duration': 0.34}, {'end': 984.809, 'text': "I'm giving it a push at the moment.", 'start': 983.328, 'duration': 1.481}, {'end': 989.973, 'text': "Oh, like I said before, if the numbers here are updating, I'm not doing anything.", 'start': 985.41, 'duration': 4.563}, {'end': 993.056, 'text': "But if they're not updating and it's moving, it's me.", 'start': 990.234, 'duration': 2.822}, {'end': 998.219, 'text': "Eventually I'll have to get some sort of keyboard camera.", 'start': 994.977, 'duration': 3.242}, {'end': 1003.462, 'text': 'You have no idea what to do with yourself.', 'start': 1002.101, 'duration': 1.361}, {'end': 1005.924, 'text': "I think we'll pull in the police here in a minute.", 'start': 1004.263, 'duration': 1.661}, {'end': 1009.866, 'text': "He's clearly no-keying way too often.", 'start': 1007.925, 'duration': 1.941}, {'end': 1015.45, 'text': "I think a police car hitting him would jolt him into action, but we'll see.", 'start': 1011.427, 'duration': 4.023}, {'end': 1023.115, 'text': "Hmm It's almost like he's doing worse than he was before.", 'start': 1015.47, 'duration': 7.645}, {'end': 1034.503, 'text': 'I wonder if this is the old..', 'start': 1024.84, 'duration': 9.663}, {'end': 1035.003, 'text': 'No, it cannot.', 'start': 1034.503, 'duration': 0.5}, {'end': 1037.243, 'text': "We've got to buy another GPU.", 'start': 1035.964, 'duration': 1.279}, {'end': 1041.346, 'text': 'Man, what a bummer.', 'start': 1040.685, 'duration': 0.661}, {'end': 1044.807, 'text': 'Clearly we need a threshold for no key.', 'start': 1042.965, 'duration': 1.842}, {'end': 1046.207, 'text': 'Get out of the road.', 'start': 1045.007, 'duration': 1.2}, {'end': 1046.946, 'text': "It's a green light.", 'start': 1046.247, 'duration': 0.699}, {'end': 1052.789, 'text': 'What is she thinking? Up, up.', 'start': 1046.967, 'duration': 5.822}], 'summary': 'Monitoring a situation for keyboard activity and considering involving the police due to excessive inactivity and need for a new gpu.', 'duration': 75.025, 'max_score': 977.764, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg977764.jpg'}, {'end': 1144.822, 'src': 'embed', 'start': 1105.941, 'weight': 6, 'content': [{'end': 1106.702, 'text': "It's something.", 'start': 1105.941, 'duration': 0.761}, {'end': 1109.484, 'text': "I'm going to throw some stars on him just to see.", 'start': 1107.162, 'duration': 2.322}, {'end': 1113.067, 'text': "Although this is when everything crashed, so we'll see how we do, but..", 'start': 1109.844, 'duration': 3.223}, {'end': 1118.477, 'text': "This is me moving, I'm just waiting until we find a police.", 'start': 1116.257, 'duration': 2.22}, {'end': 1119.938, 'text': "I think there's just one right in front of us.", 'start': 1118.537, 'duration': 1.401}, {'end': 1129.879, 'text': 'You can get arrested if you just stand here.', 'start': 1122.478, 'duration': 7.401}, {'end': 1132.46, 'text': 'You should reverse here.', 'start': 1131.58, 'duration': 0.88}, {'end': 1134.94, 'text': "I'm just curious.", 'start': 1132.48, 'duration': 2.46}, {'end': 1144.822, 'text': "Yeah, and no key is the last one, so it's about 30% that it wants to do no key.", 'start': 1139.601, 'duration': 5.221}], 'summary': 'Discussion about encountering police and a 30% failure rate for a specific action.', 'duration': 38.881, 'max_score': 1105.941, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1105941.jpg'}], 'start': 860.271, 'title': 'Graphics card troubleshooting and self-driving car testing', 'summary': 'Discusses a failed graphics card causing a game crash, impacting model testing. it also describes testing a self-driving car to establish a no-key threshold to enhance performance and safety, potentially involving the police.', 'chapters': [{'end': 974.161, 'start': 860.271, 'title': 'Graphics card troubleshooting', 'summary': 'Discusses a failed attempt at running a graphics-intensive game due to a graphics card malfunction, resulting in a crash and subsequent restart, impacting the testing of a model.', 'duration': 113.89, 'highlights': ['The graphics card malfunctioned during the attempt at running a graphics-intensive game, leading to a crash, which disrupted the testing process of a model.', 'The attempt to run the model was impacted by the lack of memory, not accounting for the demands of the game and recording, resulting in a failed attempt at running the model.', 'The speaker restarted the system to address the graphics card issue and potentially free up resources for running the model.']}, {'end': 1134.94, 'start': 977.764, 'title': 'Testing no-key threshold for self-driving car', 'summary': 'Describes testing a self-driving car with a focus on the frequency of not pressing any keys, the need for a threshold, and the potential involvement of the police, aiming to improve its performance and safety.', 'duration': 157.176, 'highlights': ["The need for a threshold to address the frequent occurrence of not pressing any keys, to improve the self-driving car's performance and safety.", 'Contemplating the involvement of the police due to the excessive no-keying behavior of the car, indicating the severity of the issue and the potential impact on public safety.', 'Considering purchasing another GPU to address the issues with the self-driving car, highlighting the technical requirements for improving its functionality.', 'Experimenting with applying stars to the car to assess its performance, demonstrating a proactive approach to evaluating and enhancing its capabilities.', "The mention of updating numbers as an indicator of the car's autonomous movement, providing insights into the monitoring process and the car's behavior."]}], 'duration': 274.669, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg860271.jpg', 'highlights': ['The graphics card malfunctioned during the attempt at running a graphics-intensive game, leading to a crash, which disrupted the testing process of a model.', 'The attempt to run the model was impacted by the lack of memory, not accounting for the demands of the game and recording, resulting in a failed attempt at running the model.', 'The speaker restarted the system to address the graphics card issue and potentially free up resources for running the model.', "The need for a threshold to address the frequent occurrence of not pressing any keys, to improve the self-driving car's performance and safety.", 'Contemplating the involvement of the police due to the excessive no-keying behavior of the car, indicating the severity of the issue and the potential impact on public safety.', 'Considering purchasing another GPU to address the issues with the self-driving car, highlighting the technical requirements for improving its functionality.', 'Experimenting with applying stars to the car to assess its performance, demonstrating a proactive approach to evaluating and enhancing its capabilities.', "The mention of updating numbers as an indicator of the car's autonomous movement, providing insights into the monitoring process and the car's behavior."]}, {'end': 1539.031, 'segs': [{'end': 1175.12, 'src': 'embed', 'start': 1139.601, 'weight': 1, 'content': [{'end': 1144.822, 'text': "Yeah, and no key is the last one, so it's about 30% that it wants to do no key.", 'start': 1139.601, 'duration': 5.221}, {'end': 1149.508, 'text': "Um, actually, well, that's interesting.", 'start': 1146.207, 'duration': 3.301}, {'end': 1155.71, 'text': 'Why is it doing no key? I think no key is the last one and, uh, I probably should have labeled these.', 'start': 1149.548, 'duration': 6.162}, {'end': 1165.853, 'text': 'Um, but no, clearly it should be doing zero, one, two, three, which might just be turn.', 'start': 1156.83, 'duration': 9.023}, {'end': 1173.456, 'text': 'Let me pull up, uh, pull up the model real quick and look at it.', 'start': 1166.814, 'duration': 6.642}, {'end': 1175.12, 'text': 'Because maybe he is.', 'start': 1174.36, 'duration': 0.76}], 'summary': 'About 30% of the model is not recognizing the last key, and there is a need to review and correct the model.', 'duration': 35.519, 'max_score': 1139.601, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1139601.jpg'}, {'end': 1293.836, 'src': 'embed', 'start': 1220.807, 'weight': 0, 'content': [{'end': 1229.232, 'text': 'So looking here, 0, 1, 2, 3 is a D, which is actually reverse, right? No.', 'start': 1220.807, 'duration': 8.425}, {'end': 1230.352, 'text': "No, that's turn.", 'start': 1229.692, 'duration': 0.66}, {'end': 1237.697, 'text': 'Yeah So he just wants to turn right in this case, which is not good.', 'start': 1230.372, 'duration': 7.325}, {'end': 1248.448, 'text': "I'm not really sure why he wants to turn just right, but okay, but he's not moving.", 'start': 1241.246, 'duration': 7.202}, {'end': 1252.59, 'text': "and I feel like if we don't do any movement for a period, that's great.", 'start': 1248.448, 'duration': 4.142}, {'end': 1256.931, 'text': 'maybe like no movement or reverse for greater than five seconds,', 'start': 1252.59, 'duration': 4.341}, {'end': 1262.793, 'text': 'probably pick the second most frequent option or just feed it some more training data.', 'start': 1256.931, 'duration': 5.862}, {'end': 1283.566, 'text': 'this is already light years better than when I fed it 20,000 data so I might change no keys to.', 'start': 1262.793, 'duration': 20.773}, {'end': 1284.888, 'text': "Ah, well, because the network's running.", 'start': 1283.566, 'duration': 1.322}, {'end': 1285.768, 'text': "But maybe what I'll do..", 'start': 1284.928, 'duration': 0.84}, {'end': 1289.572, 'text': 'I can change no key to just do forward for now.', 'start': 1285.768, 'duration': 3.804}, {'end': 1291.954, 'text': 'Because he still has reverse.', 'start': 1290.913, 'duration': 1.041}, {'end': 1293.836, 'text': 'So if he wants to reverse, he can reverse.', 'start': 1292.014, 'duration': 1.822}], 'summary': 'Analyzing data to minimize movement, considering reverse and forward options.', 'duration': 73.029, 'max_score': 1220.807, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1220807.jpg'}, {'end': 1539.031, 'src': 'embed', 'start': 1467.769, 'weight': 3, 'content': [{'end': 1470.071, 'text': 'But obviously you want to have something that can go through the city.', 'start': 1467.769, 'duration': 2.302}, {'end': 1474.436, 'text': 'And not crash into random walls.', 'start': 1473.135, 'duration': 1.301}, {'end': 1475.617, 'text': "That'd be great.", 'start': 1474.576, 'duration': 1.041}, {'end': 1500.998, 'text': "Just don't hit the freaking tree, please.", 'start': 1499.477, 'duration': 1.521}, {'end': 1504.32, 'text': 'Sorry, everybody.', 'start': 1503.699, 'duration': 0.621}, {'end': 1507.942, 'text': 'Look at him reversing.', 'start': 1506.981, 'duration': 0.961}, {'end': 1510.743, 'text': "Oh, he's getting out of the situation.", 'start': 1507.962, 'duration': 2.781}, {'end': 1512.904, 'text': "Not very well, but he's reversing.", 'start': 1510.763, 'duration': 2.141}, {'end': 1515.646, 'text': "Now he's not.", 'start': 1513.425, 'duration': 2.221}, {'end': 1517.567, 'text': 'All right, all right.', 'start': 1516.426, 'duration': 1.141}, {'end': 1519.048, 'text': "Let's see.", 'start': 1518.507, 'duration': 0.541}, {'end': 1537.148, 'text': 'Go, man, go, go, go.', 'start': 1535.506, 'duration': 1.642}, {'end': 1539.031, 'text': "OK, that's not where you want to go.", 'start': 1537.168, 'duration': 1.863}], 'summary': 'A vehicle navigating through obstacles in a city, avoiding crashes and reversing.', 'duration': 71.262, 'max_score': 1467.769, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1467769.jpg'}], 'start': 1139.601, 'title': 'Autonomous vehicle model analysis and training self-driving model', 'summary': "Includes analysis of an autonomous vehicle model, indicating a 30% preference for 'no key' action, potential issues with turning only right, and a consideration for no movement or reverse for greater than five seconds. additionally, it covers training a self-driving model with 20,000 data, encountering behavior issues, and the need for further training and adjustments.", 'chapters': [{'end': 1256.931, 'start': 1139.601, 'title': 'Autonomous vehicle model analysis', 'summary': "Discusses the analysis of an autonomous vehicle model, noting a 30% preference for 'no key' action, potential issues with turning only right, and a consideration for no movement or reverse for greater than five seconds.", 'duration': 117.33, 'highlights': ["The model shows a 30% preference for 'no key' action, indicating a significant inclination towards inaction.", 'The analysis reveals a tendency for the vehicle to primarily turn right, raising concerns about its limited range of actions and potential navigation issues.', "There is a consideration for implementing a rule where no movement or reverse for greater than five seconds would be considered optimal, suggesting a need for further exploration and potential adjustments in the model's decision-making process."]}, {'end': 1539.031, 'start': 1256.931, 'title': 'Training self-driving model', 'summary': "Discusses training a self-driving model with 20,000 data, encountering issues with the model's behavior, and the need for further training and adjustments.", 'duration': 282.1, 'highlights': ["The model's performance improved significantly after feeding it more training data, showing light years better results.", "The chapter emphasizes the need to adjust the model's behavior, such as limiting its ability to go in reverse, to prevent undesirable actions like crashing into walls or getting stuck in parking lots.", 'The speaker expresses the desire for the model to navigate through the city without crashing into random walls and highlights the importance of having a model capable of handling diverse environments, not just freeways.']}], 'duration': 399.43, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1139601.jpg', 'highlights': ["The model's performance improved significantly after feeding it more training data, showing light years better results.", "The model shows a 30% preference for 'no key' action, indicating a significant inclination towards inaction.", 'The analysis reveals a tendency for the vehicle to primarily turn right, raising concerns about its limited range of actions and potential navigation issues.', "The chapter emphasizes the need to adjust the model's behavior, such as limiting its ability to go in reverse, to prevent undesirable actions like crashing into walls or getting stuck in parking lots.", "There is a consideration for implementing a rule where no movement or reverse for greater than five seconds would be considered optimal, suggesting a need for further exploration and potential adjustments in the model's decision-making process."]}, {'end': 1837.07, 'segs': [{'end': 1675.649, 'src': 'embed', 'start': 1572.311, 'weight': 0, 'content': [{'end': 1576.154, 'text': 'It at least was making choices, which is an improvement.', 'start': 1572.311, 'duration': 3.843}, {'end': 1588.106, 'text': "So I think probably what I'm going to do is I'm going to try to just apply maybe some basic logic to this or just probably give it some more training data.", 'start': 1576.935, 'duration': 11.171}, {'end': 1589.327, 'text': 'Maybe both.', 'start': 1588.847, 'duration': 0.48}, {'end': 1592.77, 'text': "Anyway, that's all for now.", 'start': 1591.109, 'duration': 1.661}, {'end': 1594.151, 'text': "That's where I'm at at the moment.", 'start': 1592.91, 'duration': 1.241}, {'end': 1600.478, 'text': "It's basically the exact same thing as before, just a slightly larger AlexNet.", 'start': 1595.953, 'duration': 4.525}, {'end': 1602.52, 'text': 'Let me pull up AlexNet real quick just to show you.', 'start': 1600.518, 'duration': 2.002}, {'end': 1607.385, 'text': 'So this is the original AlexNet, and then I made this AlexNet, AlexNet 2.', 'start': 1602.901, 'duration': 4.484}, {'end': 1611.43, 'text': 'It has a few more convolutions in it.', 'start': 1607.385, 'duration': 4.045}, {'end': 1622.338, 'text': 'Has double the fully connected layers the 4096 layers and then before dropout was 50% I kind of I went ahead and made that 70.', 'start': 1612.731, 'duration': 9.607}, {'end': 1625.681, 'text': "I was really surprised actually I've never had success with the 50% dropout.", 'start': 1622.338, 'duration': 3.343}, {'end': 1633.026, 'text': "I didn't even realize it was at 50% before But that actually, you know, it actually did work pretty well for other models.", 'start': 1625.701, 'duration': 7.325}, {'end': 1635.946, 'text': 'So I Whatever.', 'start': 1633.126, 'duration': 2.82}, {'end': 1639.128, 'text': 'But this one just was not generalizing at all.', 'start': 1636.406, 'duration': 2.722}, {'end': 1640.789, 'text': 'It was not learning anything.', 'start': 1639.148, 'duration': 1.641}, {'end': 1642.449, 'text': 'So I changed it.', 'start': 1641.909, 'duration': 0.54}, {'end': 1645.331, 'text': "Anyway, so this one's working.", 'start': 1643.69, 'duration': 1.641}, {'end': 1646.712, 'text': "It's a very large model.", 'start': 1645.591, 'duration': 1.121}, {'end': 1650.614, 'text': 'A lot of people were asking, what am I running this on? This is on a Titan X.', 'start': 1647.032, 'duration': 3.582}, {'end': 1656.737, 'text': 'So basically 12 gigs of VRAM is what you pretty much need to run this.', 'start': 1650.614, 'duration': 6.123}, {'end': 1659.078, 'text': 'Well, or 12 gigs of RAM, I guess.', 'start': 1657.257, 'duration': 1.821}, {'end': 1665.181, 'text': 'But it would take a month probably or more to train this on a CPU with RAM.', 'start': 1659.138, 'duration': 6.043}, {'end': 1667.638, 'text': "Anyway, that's all for now.", 'start': 1666.576, 'duration': 1.062}, {'end': 1669.14, 'text': "I'm going to keep working on it.", 'start': 1667.658, 'duration': 1.482}, {'end': 1671.904, 'text': 'Hopefully get something a little more useful.', 'start': 1669.4, 'duration': 2.504}, {'end': 1674.107, 'text': 'Something that actually works.', 'start': 1672.985, 'duration': 1.122}, {'end': 1675.649, 'text': 'I think, first of all, more data.', 'start': 1674.227, 'duration': 1.422}], 'summary': 'Improved alexnet model with double the fully connected layers and 70% dropout, requiring 12gb vram to run.', 'duration': 103.338, 'max_score': 1572.311, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1572311.jpg'}, {'end': 1727.62, 'src': 'embed', 'start': 1696.857, 'weight': 4, 'content': [{'end': 1703.758, 'text': 'Instead, what I would do is reinforcement learning in the sense of the training data itself will be successful training data.', 'start': 1696.857, 'duration': 6.901}, {'end': 1708.139, 'text': "So reinforcement learning, that's basically the whole point of it, is it chooses the best path.", 'start': 1704.118, 'duration': 4.021}, {'end': 1712.363, 'text': "Whereas for us, I think it's probably better just to choose a path,", 'start': 1709, 'duration': 3.363}, {'end': 1719.711, 'text': "and there's really no good way to determine best path that I know of personally in this exact example.", 'start': 1712.363, 'duration': 7.348}, {'end': 1727.62, 'text': "So where reinforcement learning works is where you actually have a game, a single score, and like it doesn't have to be a game,", 'start': 1719.811, 'duration': 7.809}], 'summary': 'Reinforcement learning chooses the best path using successful training data.', 'duration': 30.763, 'max_score': 1696.857, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1696857.jpg'}, {'end': 1819.719, 'src': 'embed', 'start': 1797.184, 'weight': 5, 'content': [{'end': 1804.907, 'text': "And it would just kind of analyze what's going on in game to determine if these are good training data sets and then use those training data sets to further train the model.", 'start': 1797.184, 'duration': 7.723}, {'end': 1807.208, 'text': 'And then that way it trains itself over time.', 'start': 1805.207, 'duration': 2.001}, {'end': 1810.949, 'text': "It's sort of reinforcement learning, just not key learning.", 'start': 1808.248, 'duration': 2.701}, {'end': 1814.175, 'text': "Anyway, that's all for now.", 'start': 1812.614, 'duration': 1.561}, {'end': 1816.577, 'text': "I'll update y'all whenever I finish.", 'start': 1814.395, 'duration': 2.182}, {'end': 1819.719, 'text': 'Maybe add some more data to this one.', 'start': 1818.198, 'duration': 1.521}], 'summary': 'Analyzing in-game data to train model for reinforcement learning.', 'duration': 22.535, 'max_score': 1797.184, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1797184.jpg'}], 'start': 1539.051, 'title': 'Ai model evaluation and reinforcement learning', 'summary': 'Evaluates a modified alexnet 2 model running on a titan x and discusses the need for 12 gigs of vram or ram, at least 500,000 data points, and the use of reinforcement learning for training to achieve successful self-driving car algorithms.', 'chapters': [{'end': 1650.614, 'start': 1539.051, 'title': 'Evaluating alexnet 2 model', 'summary': 'Discusses the evaluation of a modified alexnet 2 model, with improved convolution and dropout layers, which resulted in better decision-making but still requires further training and logic application, running on a titan x.', 'duration': 111.563, 'highlights': ['The modified AlexNet 2 model shows improved decision-making, although it still needs more training and logic application.', 'The model has double the fully connected layers and increased dropout to 70%, leading to better performance compared to the previous 50% dropout.', 'The AlexNet 2 model is running on a Titan X, addressing the query regarding the hardware used for its operation.']}, {'end': 1837.07, 'start': 1650.614, 'title': 'Ai training and reinforcement learning', 'summary': 'Discusses the need for 12 gigs of vram or ram to run the training, the desire to have at least 500,000 data points, and the decision to use reinforcement learning in the training data itself, rather than applying q learning. it also outlines the plan to analyze the in-game performance to further train the model and the goal of achieving successful self-driving car algorithms.', 'duration': 186.456, 'highlights': ['Training requires 12 gigs of VRAM or RAM and at least 500,000 data points, with the current data at 112.', 'The decision is made to use reinforcement learning in the training data itself, rather than applying Q learning.', 'The plan is to analyze the in-game performance to further train the model and achieve successful self-driving car algorithms.']}], 'duration': 298.019, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/KSX2psajYrg/pics/KSX2psajYrg1539051.jpg', 'highlights': ['The modified AlexNet 2 model shows improved decision-making, although it still needs more training and logic application.', 'The model has double the fully connected layers and increased dropout to 70%, leading to better performance compared to the previous 50% dropout.', 'The AlexNet 2 model is running on a Titan X, addressing the query regarding the hardware used for its operation.', 'Training requires 12 gigs of VRAM or RAM and at least 500,000 data points, with the current data at 112.', 'The decision is made to use reinforcement learning in the training data itself, rather than applying Q learning.', 'The plan is to analyze the in-game performance to further train the model and achieve successful self-driving car algorithms.']}], 'highlights': ['The model uses a unique data set with nine input options and 112K training samples, but only a little over 10,000 for each key press, leading to limited key presses available for navigating the city environment.', 'The AI was trained for 13 true epochs through the entire 30-file data set, despite the display indicating 390 epochs.', "The model's performance improved significantly after feeding it more training data, showing light years better results.", "The model shows a 30% preference for 'no key' action, indicating a significant inclination towards inaction.", 'The model has double the fully connected layers and increased dropout to 70%, leading to better performance compared to the previous 50% dropout.', "The chapter emphasizes the need to adjust the model's behavior, such as limiting its ability to go in reverse, to prevent undesirable actions like crashing into walls or getting stuck in parking lots.", 'The user encounters issues with the model reversing excessively, with the initial move being to reverse 90% of the time, leading to the need for intervention.', 'The graphics card malfunctioned during the attempt at running a graphics-intensive game, leading to a crash, which disrupted the testing process of a model.', 'Experiencing frustration and technical issues while attempting to run TensorFlow 1.0, facing multiple crashes and errors.', 'The decision is made to use reinforcement learning in the training data itself, rather than applying Q learning.']}