SLICED Competition Lap 3: Live Screencast

hi i'm dave robinson and welcome to another screencast we're going to be competing in the slice competition for competitive machine learning so uh this is my third time in the slice competition it's episode nine the first round of playoffs i'm really excited about it we just got the data set five minutes ago in 10 minutes we're going to start coding so right now i'm just looking at the data set getting warmed up and it's really there i i think this is one of my favorite data sets just so far that i've seen yet the question is can you predict whether a hit is a home run based on uh some data about the the hit and this is a lot of data about uh a particular kit so you've got things like the um uh besides the id you've got time you've got teams on the home and away and which with the battery who might be playing home might be playing away batter's name pitcher's name so like very sparse categorical data but definitely important some players are better pitchers than others so i'm just going to throw that in there quickly as like uh what do i have i have um so i've started taking some notes and time series uh we have just one time series so far and we have sparse categorical i'm rearranging these just a little bit as like what might i want to uh to look like yeah so categorical sparse we got um uh the yeah that's true of battery unique id too it's like probably just gonna need one of them we don't need i really need a name and an id we have uh bully we have like boolean it's like a yes no uh with the bad of control update i know that i know that used to be important i think it's got a little bit less important for complicated reasons i actually wrote a book about baseball um about baseball statistics called um uh what is it here i'm going to advertise it because i can't write code yet phase baseball if you want to check it out we got the um uh i've got the book introduction to pure conveys performers so i'm going to advertise and i'll say for more about for more on baseball stats check out my book introduction to empirical is examples from baseball statistics so the um uh yeah i'm gonna save it uh nine and um yeah okay so let's keep breaking this down a little bit so there's like ideas is boring and the idea is just like um kind of nothing i'm gonna skip it do you have a categorical uh categorical less sparse that ball classification what is a batted ball classification uh it's um let's see bb type ooh this is so cool it's like almost it's like normal data bait roughly normal on played x and z you can do a heat map uh pitch miles per hour launch speed launch angle that's gonna have a lot of impact on whether something's a home run you can you better believe that and uh yeah we got like strike um strikes falls out so that where was it in in the um where in the game was it inning outs strikes falls we're going to call game state i want to bring my numeric ones up to the top i don't say that so yeah definitely some like heat map kind of things the um let's see bearing is going to be not so i think oh let's see home away batter name etc etc etc bb type and bearing right center left and other so right left center so that goes in the category not not so sparse we have bearing and pitch name i think is not as sparse we have a park id that's what's really cool is we have um this is going to be pretty cool we actually have a data set to join this with let's say game out yeah yeah a game state here we go yeah i think this covers reserve and then park uh well what do i have about each part i've got the oh look at this all six distances okay this is kind of oh okay this is really cool we got like left field wall center field wall right field there's just a real wealth of like data uh yeah i'm um i'm really excited about this so the seo cover let's check cover this is what wow there's gonna be one of my favorite all right so cover is um home you see categorical sparse cover all right and the um and we also have a bunch of numeric ones we've got so much you can do here that's really cool okay like wow this is a really rich data set yeah the um ideas let's see examine numeric first i think like i really just i really want to see like these get a summarize home run function and then like hmm i can do it in a lot of ways like i can kind of um uh yeah i can bucket these that might be the way to do it all right this is gonna be i think yeah i think it's really cool join with stadium off the back so like or park i should say okay and left yes we type bearing pitch name yeah this is gonna be really exciting i'm pumped yeah i'm um i'm pumped uh let's see okay yeah just wait until nine o'clock i'm gonna miss i'm just gonna join together time oh time says hmm all right the other fun thing about today is that i'm doing some i've loaded up a couple memes and i'm gonna try math so the me so one thing that the traditional slices like show memes especially as a model is plotting and i've kept a few that uh that i can load up and i'm excited so let's do this library tivers library um library models oh uh yep we're coding and library stacks and i'm not going to text recipes but i'm going to do theme set theme lite i'm also going to do library lubricative i'm also going to do library meme so the fun thing about me is that i can say i can do this i can do i can bring in some information from the meme templates that i've done so uh for instance the there we go oh yeah let's see this works so the um so this is generally like like uh the story of me and lana who's one of the the competitors who has been just doing fantastically in the slides but i think that my chip having the chat vote i can do pretty well uh uh do parallel register do parallel chords equals four and the um uh yeah so let me see i'm missing something yeah uh what i do is bring in my training data train i'm gonna do um data set is read csv downloads tag sliced episode nine uh train uh and um i'm gonna do part and uh left join parks by what was it called um i missed this it was part it was parts with what park id part nope just the part and uh yeah i'll do hold out is that csv and maybe one other thing i'm going to say in uk is home run equals factor if else is home run yes no so just uh get that into a um and what i'm going to do is i'm going to build a little function called summarize h home run function and i always do something like this where i say summarize and home run is sum is home run equals yes pcp home run is hr over m so i can take my data set oh and i am oh i almost forgot split initial split is a data set 25 and 25. how many observations do i have in the data yeah the 46 this is going to be great the um and training split test is testing split and the up i forgot to set seed 2021 train the train five-fold is framed before i forget f5 yeah here we go and the um all this this data and uh yeah i've got it by um yeah i've got to buy a bunch of things so i can start finding except finding stuff out so i'm gonna do i'm also gonna rename park named his name i don't like that i don't like that uh that name right there wow i have so many things i can do and i'm gonna start with parts so i'm gonna say summarize home run and get the low and high confidence bounds says q beta 0.025 of um hr plus 0.5 uh and plus point n minus hr plus 0.5 pi is q beta 975 uh hr plus 0.5 and minus plus point five in my confidence bounds and pct hr um park name error bar h and i need a few more steps here i'm going to say part name is fct reorder part name by pcphr i'm gonna need an aes x-men is low y x max is high scale x here's labels equals percent new yankee stadium is easy to get at home running wow look at that uh and it's uh the yankees win the uh let's see we got the uh wide park name what is great american ballpark what is that oh i got the home team name uh i can actually do um yeah uh brave american ballpark the cincinnati reds okay yeah the um it's easier to hit a home run in the great american ballpark or a new am subtitle is um and i'm going to do a few little tricks to say sizes and uh let's see the um is 95 confidence interval okay okay yeah okay and uh size is number of kits okay the um yep do you think that's a golden feature not so sure so i've got i've got a thought on this which is the um uh which is generally the um [Music] nick i think is pretty challenging in terms of like uh finding golden um specified gold features you can find one that i can't find but uh yeah let's see how we um how we do i'm gonna do year equals year of um uh year of game date which is game yep game date and yeah gear heart up pc phr size is good um here it should be y min y max why not this y year how has it been trending overall uh oops the uh oh it's only one year all right then that actually makes it a lot easier this is in one season it's another pandemic season uh then let's try month set game date um i didn't realize it was over a small period of time postseason post season matters i'm going to say a few group equals one yeah the um postseason matters uh you guys will in future we'll find out if that was a golden feature uh but uh yeah i'm gonna leave in my my name means i think i will but yeah the um see so postseason matters i'm going to stick to that and uh the um oh yeah so oh yeah so let's look at um let's look at park effects such as i'm going to do that a little bit earlier and i'm going to say park effects let's see data set um you know i'm going to put my part name i'm going to put my other data department uh i'm going to group by park name and um let's see uh left yeah let's see i'll all do this lf dim cf dim uh you know i don't love the um the naming here i'm actually going to bring it take these parts and do janitor clean names right off the bat and uh um oh that's gonna mess up a different different step clean names oh uh yeah no it's not gonna mess with this stuff oh i've been using data set i should have been using training technically i should have been using my training data a subset of data um and uh uh and here yeah now i'm a bit happier i left him ah lft is actually not really an easy shortcut for this like you can't use your selectors and by part uh let's see lm pct hr explained by lf dem plus cf kim plus rf dim uh the um i just want to see like a a uh uh data equals dot so let's see no right now i'm not seeing anything yet out of the dimensions uh and the um yeah what is the the let me see the field dimensions so lf distances and height uh so if i try like cf dim plus cf height uh cf 8 cfw wall like for all yeah i'm just not i'm not seeing like something like like popping up yet when i do like um cf w uh by uh pcp hr yeah um well there's like a lot of noise there but it is worth noticing that like uh if i do jiggy repel um name actually i'm not going to do that i'm going to do check overlap if it's true and you know do any of these correlate like at all where if i just did like gather um from lf dim through wf uh rfw uh if i say like mets metric value or equals list for that test of value with pct hr i'm doing a little test unless um um and i do make this attie actually i can just do heidi chord.tess i think i can do this let's find out what happens yeah there it goes the um uh no significant p values but that's just that's on the fraction i should really do it they're not independent other ways but yeah okay there's a few ways i can do that but yeah i'm not going to use i'm not going to um i'm going gonna bring it in as as training things but it's not huge let's see let's see the um angle and let's use the mirror all right uh so let's see um in the meantime i'm gonna do meme uh memes and uh let's see so um oh yeah so look at this if i do fill equals is home run yeah i'm going to want a gm pilot a heat map for some like x and z and look at that launch speed really predictive and the um and i need to make this alpha equals 0.5 uh there you go uh oh look at that watch angle huh there is a few observations there's a sweet spot of launch angle home runs happen uh of launch angle and launch speed triangle midway gave away and launch speed the high you know let me try to get that into a heat map and i'm gonna do uh yeah those are my two favorites so i'm gonna do a group by let's see um uh bin is pile uh of um this i haven't used before but it's like launch angle uh launch angle bucket launch speed bucket launch speed and uh oh we do summarize tcp home run is mean is home run is yes but also end and also min launch and or the average launch let's try to decide like what's the way to do this no this was to this is going to be yeah what i'm going to do is round to the nearest ten around minus one and round to the nearest the world's biggest fan of but a lot and i'll do filter ks launch speed bucket launch angle bucket and fill equal pcthr and uh wow look at that i'm gonna [Music] turn up the resolution a little it doesn't care for that yeah what is it zero let me see yeah i need to have some rounding labels equals um percent oh no it's a radiant two [Music] mm-hmm why's it got so much empty space over here i thought i filtered those um that's so weird oh i get it i just need to get rid of all things all right oh this is launch speed launch angle all right there let's see yeah um make it a little lower okay angle high speed wow that's really cool okay and um high speed okay uh anything else i want to do here i want to say um subtitle yeah i multiply it by two yeah okay no that's not true that around to the nearest two so there's five on each scale and um no buckets only with less than 30 points okay and um you know now that i've done that i want to look at my other one large speed launch angle pitch x played x plate z that that like fits really naturally played x plate z see i'm tired of hyping bucket i wonder no the rounding will need to be the nearest harmless place at least and um said xbox z tr x it's c man the uh uh so if i do i need to coarsen this even a little bit ooh wait yeah what if i did i change the scale before i do that and i say point one um [Music] yeah then let me say like what's the bullseye what is how does this describe to the ball position center plate home plate place up distance you know realize notice that like x-intercept 0 is special here so like generally percent scale percent where did i go where did i go oh right yeah uh labels oh yeah it's going to be fill equals percentage that are home run uh five all right these are some good uh these are some useful numbers and he did not look as important all right it just occurred to me dad i said a metric yet okay i'm gonna do a couple other things uh while i'm uh well i am let's consider uh let's see uh the um so i like about static statistics isn't interpreting my model but um but i'm gonna get i'm gonna get back to that metric set and uh um i mean log loss control is grid control control grid save cred true save workflow true i'm going to start fitting some models and um keep an eye on okay so that train um um let's see uh pitch and ph launch speed launch angle uh plate x plate z [Music] you know i didn't um look at my other other numeric ones game state i want to see if it's if it's very important it's very interpretable so i'm going to start with like training so it looks like maybe games they're a little less likely let's see any percentage home run and uh yeah we're on that i'm not going to incl i'm going to do actually so i'm going to say indian is if else now i'm going to do keymats human in pen so um because this is the 10 plus the postseason so you say okay there's a slight trend slight trend i'm not going to change this right here on the ribbon alpha equal point yep okay home runs just wanted something quick and uh let's see can i find anything else with the game state the um state meaning outs went up it's uh zero one two it's really easy not really uh out i don't know yeah balls hey um more like me oh you know what makes sense here balls matter because people are a little more caught picture's a little more cautious there's a more center and uh i bet that when the ball uh i bet the three balls that uh oh you know what i can do i can actually do like a group by balls and spikes to get a little grid yeah strikes not to do by out because it looks like out doesn't really matter and make this a little pile uh huh no strikes fail equals piece hr scale fill continuous labels equals sale fill low is blue high is red midpoint is 0.08 looks like labels are both percent any better it's that white the exact white is bugging me a little yeah let me let me make this little sign there now x is a number of balls number of strikes and i'm gonna put that in this title three three balls two strikes or close you know i can actually say i can change this a little bit and say does this because they tend to be more in the center so i can actually ask like summarize pcphr is mean we do a little causing understand the mechanism is this because fatter pictures are more likely to aim for the center of the plate with a full count average plate uh absolute type c mean absolutely full count no not really uh no not really look average absolute plate z pc hr yeah okay actually yeah okay a count is a page zero of uh balls strikes no um wait i did something off oh with oops not a full count sorry for that the point is you have three balls no strikes you're really uh cautious and yeah it's like the 3-0 and the three one up average oh uh oh z average height this is height which doesn't need to be absolute average absolute distance center watch this uh here we go okay and give me a moment to grab this same situation [Music] see why is it not um oh it's not yet uh the means you could create a feature of balls minus strikes oh um center plate this is cool but it actually means it's not that important once we oops all right so let's fit an extra uh on the uh on the numeric variables and why not do any ball strikes doesn't hurt uh really yeah it doesn't hurt like you know it's all just um and is pitch oh i forgot his picture lefty yeah i'll find out here now i forgot to do one that i think is going to be really important really boss and that's grouped by uh uh db type i think this is sort of data leakage because um uh because like yeah the um uh i'll just do yeah okay or not isn't a dvd pipe yeah ground balls and pop-ups yeah i was curious about this one and we definitely need to include it i mean it's like um it's not gonna uh it's not gonna hurt anything to get a scale like uh x's yeah um yeah literally never uh are literally never home runs all right so that means i'm definitely not going to go without c including baby type and uh yeah all these other ones i'm going to drop any ball strikes those were a little bit weaker and i'm going to find out if daddy left the et cetera is any good okay so uh uh let's see and um work i'm gonna pipe it into workflow f as um boost boost tree classification oh i did dot what am i doing uh is home run uh learn radio start with a fixed learn rate because i have a habit i like which is the i'm try is tune trees is uh about control equals yeah grid is crossing um i'm try i don't know two three four trees is 300 plus 700 by 100 and the first argument train first i'm gonna throw up a meme what am i going to do the uh uh yeah i'm basically storming in some default parameters so this is the mean but that's about uh but yeah and um 25 all right the uh oh xp tune and while i'm at it i'm also gonna write up a little code for xg uh xp finalized is xp oh oh that calls for a meme oh boy the uh calls for a meme but uh unfortunately let's see where what i'm missing oh i see what i'm missing i didn't do step dummy i've got a guess on this the uh uh-huh yeah yeah this always happens step dummy all nominal predictors i completely neglected to uh to include that um yes the story is like i have a nominal predictor they're up and i actually have numeric and i'm gonna want to step these mean a lot but i'm doing median i'm an acute median of numeric predictors stuff unknown all nominal oh um yeah the um the notes column has been really a kind of a joy to mess with um but yeah i'm going to try i'm just going to like get a feel for how it is actually tune plan tune uh i'll add mirror columns and less sparse categorical true and hyper get then go to uh then go to the all right uh yeah so let's see while this video i'm also gonna save i forgot to save a lot uh see and uh i really do have to load this and uh say uh f9 have my eda ver i have my eda notebook open memes are nice but but being able to spend that time your uda was even better uh huh i think if it was going to fail to fail now it's pretty big that's why i like the train five-fold it's taking its time um it's got a lot of data in there it means that maybe i should have started fitting these a little bit earlier but yeah um kind of i'm just gonna copy this whole thing into here and say idiot let's see uh right i'm gonna look a little games and players so i have a stadium analysis up earlier where i can i look a little bit like categorical data i kind of want to go in from that example that paradigm and still do that uh of want to go from that example and say uh let's make this show a categorical categorical oh i'm not going to do it on these based on comfortable [Music] and [Music] so i usually means i miss the i'm using call incorrectly here somewhere up oh yeah whoa all right and let's see what i got mean log loss oh ah oh i don't have mean log loss accuracy got better auc got better it was way better by 800 ah ah i just was really um all right so i know i don't want to randomly yeah i'm going to try four and five and you know kind of want to just like yeah try to be a little fast with this all right so i've got this like this paradigm uh that is this example of how to solve it and i can do theme actually uh summarize clock category i might need to change this i'm gonna move all this code down here oh yeah i got a meme for that i noticed i gonna kinda gone a while without saving so i can actually uh yeah i mean for that all right uh oh uh yeah so that's the batters i want to the batters team and the uh team batter team and if i say yeah oh here we go let's run uh okay so five is better all right that's five is better enough into five six seven i don't think i need to go farther than this but i am point and uh collect mean log loss uh that's uh um see i don't keep this stuff oh one that i forgot is time series what's the postseason uh see i want like laid up let me see is october okay yeah basically later the postseason the better is that is that anywhere in the data like because look it actually does get a lot more likely that could be explained by other things i'm just going to add this postseason for now it's a simple one and say step mute is post it's october all right what am i doing right now i wanted to um crap juice i want to just check uh-huh oh i forgot game date and um you know it'll find some of the breaks itself so i'm not going to do i'm not going to do is october it's going to know it's going to find that i'm just going to do step youtube game day people at um day day of game date and remind yourself that day of systopic is the day within nope that's uh i'll do a week just uh yeah sure why not week within the year benin falls strikes and uh six seven i mean ball strikes what else did i have that was uh pretty good m kitchen ph but all these and if i did cover thing i don't think it's gonna cover summarize home run dome a little higher but that's not bad that's nothing and they all kind of overlap um yeah that hover doesn't really matter so i'm just gonna yeah i'm gonna go ahead and do five six seven eight i guess in case i needed i got a little farther on this all right and uh well that runs uh so while that runs i'm going to yeah i'm going to look at batakin [Music] and uh this time is this category i don't use i didn't bring in cap blue or haven't had practice and cap boost but let's try some of these um who's different percentage which are the great season others less so this is cleveland who i think is called the guardian st louis cardinals the dodgers the chicago white sox uh it hits oh no uh yeah so team is going to be one of those categorical ones uh you know i could turn it numerical and the um the the boot x2 boost might find uh like trends i really want to like order it but i it's not really it's kind of about cheating to uh order it i still might do that yeah what i'm gonna do yeah i'm gonna do this it's not cheating it's like um it's keen ordering is fcp reorder uh oh you're only on train data try turning uh st reorder uh train uh team home team batter team ordering sure i should have added like this home that's an important one uh home team train and um park ordering as ordering i'm only using the training data but it is going to make it look unusually good but but it can always find interesting um um levels train park nope something's up it's still alphabetical and that's to reorder oh mean orders within their app based on their average okay and the cleave at the bottom uh near kinky's top okay yep all right the um oh let's see what we got okay six kind of worked out the best okay i'm try six uh-huh let me try six six and a few more predictors all right i'm only going to try i'm adding creatures now so i'm going to add snap mutations home is home equals home team let me check some real quick uh okay um yeah it matters a little it matters a little i'm gonna do a new plot i'm gonna say is home and you know i'm just i'm going to add it because step is home is home and we're going to add [Music] i'm trying to turn the [Music] all right um oh park was not included here heart name park name park name before i do this let me bake it real quick and say so uh juice nope it failed home up oh select yep well um really important need to get rid of that categorical variable yeah this is pretty good um is home team where is his home his home let me get his home [Music] all right well that runs we go back here we're like okay let's um oh here's home let's see where i got on this uh huh um oops we'll see if it ends up being important i'm gonna um i'm gonna go a little farther and add a uh and add a finalize parameters finalize workflow select best sg tune let's do workflow finalize sinusoidal just actually fit this actually workflow finalized fit frame so my channel all right now i add all that had all these things and it's like hard to get any better let's see hmm and use of the lefty i use ah bearing baron i forgot a baron all right now uh makes sense that this fits whatever we were seeing like yeah in terms of importance um bearing let's take a look at bearing all right and same thing hey bearing really matters center field is more likely i'm going to throw that into the model ah bearing let's see [Music] all right yeah we need more more we've added a bunch of things 68 10 12. this is maxing out here so i think i need to go past 800. no memes it runs i'm gonna run to the bathroom and i'll see uh let's see the um uh yeah i'll do um one more i had this once before but i'm gonna pop i'm gonna leave it up again just in case in case you missed it all right [Music] [Music] all right still training and i'm gonna do yeah so what i'm gonna do is um try a linear ma model it's gonna be a wreck and the things that we are looking at that are stars uh flat picture never got something what is that ah too sparse all right so i don't mind that i did i skipped it um yeah a little interesting but i'm going to try this as a linear model uh regular up here you go up ah next with 12 eight seems to be the best eight nine five yeah that's as good as i think it's as good as we're probably gonna get but i think um i'm gonna add no i'm not gonna add 900 and i'm gonna do train five-fold instead and yeah i'm not burying i'm gonna add not outs not pitch maybe that's parsley yeah give it one more shot with five-fold training this time okay and um [Music] oh yeah okay so the recipe is gonna be um uh all right it's gonna be is home run explained by home via let's see home team plus away team plus uh i'm gonna do better team they're all they all can be included and i do i'm gonna include the really important um db type because otherwise it's like uh huh and that um plus is let's see where uh name and i'm going to include launch speed no not launched all right step dummy nominal predictors let's start just with one value all right is uh-huh fit so hmm oh [Music] hmm i'm trying to like make sure i'm doing it just like i'm just getting a hello world all right well let's see this all right that's as good as we're gonna get uh okay let's let's try it 0.895 and then sorry [Music] um all right nine three seven okay and check yeah and his home run all right all right six percent that look yup looks like okay let's try our first attempt and let's see what was this first account all right where's uh public leaderboard doesn't mean all that much but i right under my uh yeah just on that one attempt right under uh one of my competitors and still doing all right at least but that doesn't matter that doesn't actually matter what really matters is let's see because after all um uh the uh uh big danger of this is that things look good on the public leaderboard but they don't turn up to they don't work out for us do they uh but um yeah let's do a um let's do that uh linear model in here and see if we have any more love it's just like just be stepper no no nominal step other nominal step down and then add this a lot of people just as to how it's so slow is home run logistic re oh okay that's why oh yeah that would do it that would absolutely do it um i wasn't telling it to use the generic language which is why i didn't find the penalty yeah okay now let's try okay um a little bit better than this um um oh shoot it won't be able to fit it's not it won't be able to i'm trying to do a stack then of the um i'm gonna kill this because i think it's too slow for a reason and i need to hear it on five-fold too many things got included once i know three yeah okay and fill the parameters a this is slower than i anticipated it plus 1.03 to 10 combinations well i'm going to try a little more exploration first while i'm doing actually i'm going to yeah i'm going to do a little more exploration okay uh so yeah goodbye picture name i've got an idea um i can do summarize and uh hr is some is home run it's yes n is n but he also has a few pictures of particular personalities in terms of where they go on the plate so i can say plate average plate x is mean plate x average plate y is mean plate y wow this is slow and uh average plate x average plate z and yeah let's put the color as being so i'm going to do is i'm going to say aas color is this and scale color gradient to low is blue high is red yeah i'm just going to um start with this so slow [Music] and this is the stacks package for fitting um for fitting models but yeah here we go and i'm gonna say uh point is 0.05 it's pretty similar to what it is overall and uh labels equals percent plate average height hmm let's see values that's a bummer second is so much slower than i wish it was but i'll let everyone that right that go on really we should speed that up a little bit it's only doing two predictions uh but yeah the um might be able to uh okay and i'm gonna do size mhm foreign this is just this is just you know i can do like um i can take memes and say uh let's see practice templates let me do this where i say uh it's actually the meme package lets you uh oh yeah let's see first let's take this oh but uh yeah you know something else i can do uh i've got these two um uh i've got these like i've got these two plots right i got these on this uh launch angle plot i've got this plate plot and i can try this i can do is i can say as image function uh plot and say uh file i just want to turn this into a an image object all right but first i'm going to uh get on test mm-hmm that's worse it's worse oh man all right let's fit um boost um all right all right i'll try a different kind of model after my attempt one and say uh let's do this i don't know as much about what this is gonna be so i'm gonna try a few levels six eight ten six and ten by the way that's two actors in case i don't use this uh so what i'm going to do is say did you save key and uh image read the reason for that is so i can do launch image magic lets me combine images uh-huh uh see all right and energy but 1200 while i like this i like this because these two plots i think one of them is better i originally expected uh originally expected nope oh i'll get back to that in a sec oh it's doing well already it's doing well uh six a bit better so i'm gonna do by that point so six and eight and twelve hundred thirteen hundred okay and uh yes i'm gonna do image so i'm gonna do image read um here we go combine them and i'm going to image scale height equals huge scale geometry area height is 1200 and let me see because confined is height sixteen hundred see i got this idea that maybe i can try and get um we can try and plot this yeah there you go i just need to change a couple things here launch angle plot it's not working yet oh um the title is not gonna work labs all right and here we go six and yeah it's pretty good okay then i'm gonna do uh i'm gonna stack these two [Music] stack um and oh shoot fivefold okay i'm gonna change this to fivefold this just to six and yeah the um i'm gonna do a quick thing on this labs x equals click and i'm gonna do as image 1000 it's a little better why it's still good yeah i like my um i like this as i can do yeah i can do see ah up all right up i'm just thinking about like flats [Music] and oh [Music] all right let's look one more time at all the things that we created i think that the uh launch angle like the really big one was things like the launch angle where i have that yup uh things like the launcher angle things like and remember please do go over to nick rand's twitch account and watch this um and uh do vote be sure to vote for me uh i really do like this conclusion the this one and i say the launch angle plot really kind of landed the home runs the everything and i also like this conclusion about uh the greater the number of the balls the this grid uh conclusion yeah percentage hr yeah i'd say go over oh yeah and um i like that conclusion i like this kind of like causal reasoning uh we found the reason closer to center and a bit higher falls and yeah so the um uh and then we found a lot of other things we did like we did by team we did buy uh by stadium we really did it we didn't look much like players i no player popped up that much and we discovered that linear model wasn't really that useful we saw a bunch of memes we saw the uh the let me see uh my favorite was probably the um the one on the notes and yeah which i definitely did spend my fair share of time looking at today but yeah uh all right uh head over to granite it'll be a lot of fun it's really good let's see how it turned out

Channel: David Robinson

Views: 1,788

Rating: 5 out of 5

Keywords:

Id: Ep8OGhrSAhU

Length: 128min 36sec (7716 seconds)

Published: Tue Jul 27 2021