Analyze videos of your gameplay

Thread in 'Strategy' started by EnchantressOfNumbers, 21 Apr 2016.

  1. I did some analysis of my TGM 1 PB video and wanted to share some of my findings here. This was the first time I've really gone this in depth with it and I found it really eye opening. I encourage you to do the same.

    Basically, I tracked through my video noting section times, line clears, and level stops (VLC frame advance helped me not lose my sanity with this) and I compiled all this data into a spreadsheet. Here's a screen shot of the data:

    TGM1-GM-113031-Analysis.png

    I truncated the times to tenths of a second as I figured that was good enough. I defined the level stop start as when the first piece on level x99 locked down, so if the first piece at level x99 cleared a line, the level stop was 0 seconds. Looking at this you can probably notice a number of things, but let me point some things out.
    1. I clear way more singles in the second half (20g) than I do in the first half (pre-20g) - almost 4 times as many
    2. I clear way more tetrises in the first half than I do the second half - over 5 times as many
    3. Level stops weren't a big factor in this game
    Since the line clear animation takes 41 frames (at 60 frames/second, that's about 0.68 seconds), clearing 4 singles takes 164 frames (about 2.73 seconds) - 123 more frames (2.05 seconds) than clearing 1 tetris. This time adds up: by my calculations if I had cleared 14 tetrises and 1 triple instead of the 59 singles (all other things being the same), I would have shaved about 30 seconds off my time.

    Since I found the Avoid the Single thread, I've been consciously trying to reduce the number of singles I clear when I play. I think this has paid off in the first half of the game, but clearly I need to stack better in 20g.

    Even though level stops weren't a big factor in this game, I know that they sometimes are for me. Maybe I got lucky on this game. Maybe I'm getting better about level stops. I really need to analyze more games to get a better idea about this.

    I think one of the best things about this exercise is that it can challenge the assumptions you have about your play. Going into this, I kind of expected to find that my 20g was better than my pre-20g play, but now I see that I have a lot more "free" time in 20g than I realized. I also expected to have had a lot more time spent at level stops. I know this analysis has its limitations - it doesn't take into account placement decisions, finesse, hesitation (e.g. not charging DAS during ARE or changing your mind on a placement), etc. The analysis takes some time, but I think it's worth it.

    It looks like I'll be spending more time practicing with the 20g code.

    If you're wanting to improve your times, post your own analysis of your game(s) or point out things you notice from other people.
     
    Kitaru, GyRo, Jayce and 4 others like this.
  2. [​IMG]

    I decided to do the same for my own TGM1 PB (GM @ 09:58:78). I knew my PB game was a bit sloppy due to nervousness and misdrops, but I didn't realize how few tetrises I got in the second half. I also considered myself pretty good at low-G stacking too, and, at least for this game, it looks like the opposite!

    Early on when I started grinding for sub-10, my early game strategy revolved around making a two-wide well halfway up the playfield. This allowed me to play at a moderate (not quite optimal, but still very good) pace with my stack height, but with added safety since I could also skim cleanly with most pieces. Aside from that though, I didn't realize how much I rely on doubles for stability throughout the entire game. I should put higher standards on my line clearing past level 500 instead of settling for doubles.

    I'm glad I was able to clutch out the last section. I tried extra hard to build for tetrises and I don't think I would've made my goal otherwise. Since I covered up the timer when I played this game, I'm extra glad I made sub-10 instead of 10:00:XX. Uncovering the timer to see that would've been so disheartening.

    It's really late, it was right of me to log my "LS Start" (Level Stop Start) times as the frame when I reached level X99 / 998, right? In any case, it looks like I almost never skip over those levels. Easy time to save!

    I thought my PB was going to be very difficult for me to surpass without going insane, but just looking at this makes it all seem possible again.

    @EnchantressOfNumbers, comparing our line clears and level stops, it looks like we're pretty even, but it seems like I just save so much more time per piece. This seems to reflect what I've been saying about the hesitation and uneven rhythm in your play .
     
    EnchantressOfNumbers likes this.
  3. I started counting the Level Stop Start from when the first piece of level x99/998 locked down if it didn't clear a line. I figured that you're not really losing time if you don't spend more than 1 piece at level x99/998.

    This is actually pretty eye opening for me as well. I'm not quite sure how to fix that, but I have been trying to avoid the hesitation in my play.
     
  4. I too did an analysis of my TGM1 PB run, 12:11:50.
    [​IMG]

    This certainly was eye-opening. It FELT like a relatively clean game but I guess that is what happens when you clear 91 singles over the course of the game!
    Level stops have always been a huge problem of mine but it wasn't until I did this that I understood the degree to which they are dragging down my time. Almost every second of level stop can be eliminated, and if I had done so, I'd be at 11:36. Holy shit. @KevinDDR pointed it out in the chat during the game, I had multiple 5+ second level stops, but 34 seconds is fucking obscene. Even cutting my level stop time in half would have been enough to get me under 12 minutes.

    Singles are also a HUGE problem, and until just now I didn't realize the extent to which I relied on them. Way too many singles.
    On a related note, part of the reason I had so many singles in the 100-200 section is the result of me being unwilling to clear down, which is probably a mistake. Does anyone know the best balance between fall time/clearing lines?

    I highly advise that everyone else take the time to do this, it truly will reveal issues in your play that you never noticed before. I will probably try to grind away my issues with singles (particularly in 20g, but also in low g) and level stops and revisit this thread at a later date to try and identify more precise issues with my play.
     
  5. I did this again on my new pb:

    TGM1-GM-105556-Analysis.png

    I'm not sure that it's as eye opening this time, but I think there is still some good stuff here. I was able to cut some time out of my Level Stop. I was definitely focusing mostly on correcting my uneven rhythm / hesitation that @MaryHadALittle has pointed out to me, so I cleared 10 more singles and 5 fewer tetrises on this pb than on my previous one. Even with the additional time loss to single clears, I was able to remove a significant amount of time just by focusing on placing pieces as quickly as I could. I think my next step is to get back to trying to avoid singles while still doing avoiding the uneven rhythm / hesitation I've been doing in the past.

    Edit: to include a link to my new pb video.
     
    Last edited: 29 Apr 2016
  6. spreadsheet.png

    Did an analysis of my recent TGM1 PB while at work to help pass the time.
    This run was substantially faster than the 12:11 that I analyzed previously in this thread. I did some things well and some things poorly. If you watch the video it will be glaringly obvious that it went off the rails in the 800's and I barely finished the 900's, but it was still a PB.

    First off, level stop time cut down from 34 seconds to 12 seconds. The level stop at the end of 700 was caused by what seemed like a mistake in my execution, unfortunate timing but I'm doing much better overall at paying attention to the level counter.

    Singles down from 91 to 79, which is still WAY too many. It's very obvious that the last few sections hurt in that regard, but even 600-700 was high in singles (wasn't clean either). No Tetrises in the last 2 sections. I feel like this run was carried by the first few sections and locking speed alone, if you showed me this spreadsheet without any times, I wouldn't think this was a sub-11:30.

    I need to work on my stacking and avoiding singles.

    As with before, this was quite revealing, and you should do it too. I know what I need to work on, where I improved, and even have updated time benchmarks to know how my game is progressing.
     
  7. I thought I'd give this a go on my latest PB, very interesting process; I'm not entirely sure about the level stop recording (I believe I did as Enchantress suggested). Oddly, if I've done this correctly, my game looked like it should have been much faster time wise:

    TGM_Analysis.png

    I'm pretty happy with both my tetris count, and my level stops (unusually good game in that respect). I was expecting to see a bad section that slowed me up, but actually it looks like I need to shave off time in general.

    Watching the replay back, I realise my stack was low for a lot of the game, and also I spent a lot of time maneuvering pieces (e.g. walking them down the stack, triple rotates). My stack wasn't always particularly flat/clean, probably causing the former. Looks like I have some high stack early game/20g movement practice to be getting on with =o.

    I'll try and upload this game soonish, there might be some other terrible habit I'm missing.

    Edit: Added VOD (Fairly poor quality as it's a quick recording replay).
     
    Last edited: 31 Aug 2016
  8. After talking about some of this stuff with @JBroms, I decided to do a similar analysis of 777's best recorded game - 8:48:90 (video):

    TGM1-777-Analysis.png

    The first thing I noticed was just how ridiculously fast his section times are. The only real hiccup seems to be the 400-500 section with only 1 tetris and 1 triple, otherwise from 300-999 the sections are 49 to 51 seconds each. Also there is no level stop, 777 skips all the x99 levels or clears a line with the first piece on level x99.

    777 does a great job at clearing a lot of tetrises - twice (or more) any other individual line clear type.

    The most interesting thing for me is that there is evidence that 777 could still improve the World Record by multiple seconds without the hiccups in the 400-500 section.
     
    Last edited: 1 Sep 2016
    Jayce and JBroms like this.
  9. [​IMG]
    Since 777 has been done, I decided to do a top western player to see how it compares. I chose to do @KevinDDR's 9:39:03 video because Kevin is actively playing (in Japan!) and might be able to benefit from having an analysis done.

    My take on it:
    First off, very little level stop. The one second at the end of the 300's looks like the result of a misdrop (one piece)

    9 Singles in the 900's, a few of which I feel weren't necessary. While it is best to clear the game with as low of a stack as possible, doing so at the cost of singles doesn't seem to be the way to go, and some of the singles were done with a very low stack. He also seemed pressured at the time so maybe it was just a couple of bad reads under pressure and I'm reading too far into it.

    Kevin sure made lemonade out of the 600's section.

    A couple of notes that I made along the way:

    at 4:15 (game time) he cleared a double with an S, despite having a very clean stack and not being near the top. The more efficient play would have been to place it somewhere around column 3 and use the next piece, a T, to fill in the gap.
    Same deal at 6:40, but he had just recovered from a misdrop and probably was a bit on edge. At least, I would have done the same thing.

    The single cleared from a Z placement at 5:26 definitely had another possible placement. Whether or not that was optimal, is up for debate. The same thing was done at 7:55 but in that case it was DEFINITELY the correct play.
     
    Last edited: 7 Sep 2016
    Jayce likes this.
  10. Maybe I add a small detail to this comparision. Line clear delays aside I estimate that 777 places every piece around 2 frames faster that Kevin by average. That makes up the majority of the difference but your tables do not say anything about the reasons of this delay. I would be very interesting to have a a histogram of active frames by piece, but that cannot be done manually.
     
  11. For what it's worth, both times when I smashed my PB the other day I took your comment into consideration. Another thing that helps is being aware of the "hold to lock" placement, where you just IRS a piece and hold down before it comes out to instantly lock it. It can seem a little scary but hey, 2-3 frames off of what can easily be 30-40 placements in a game is still something! TGM1 is frustrating to frame perfect lock in because it has an extra frame or two of lag over TAP, making it a little sluggish feeling.
     
    simonlc likes this.
  12. I just saw this thread today, and wouldn't you know it, I just got a PB today, too! Let's analyze it!

    upload_2016-9-16_18-12-13.png

    From what I can tell, my level stops were pretty good. I still have a big problem with singles, especially in not-quite-20G, and of course 20G itself. Had a few misdrops that really killed my time in the 300-400 and 400-500 sections, and kinda blew it in the final section, too. A shame, too. I feel like I'm ALMOST there, where I can get a sub-12 with a perfect run.

    I feel like with this run, compared to my previous PB, I was playing a bit slower and more deliberately, as opposed to all-out for speed. If anyone has any other comments/suggestions, I'd be glad to hear it.

     
    Last edited: 18 Sep 2016
    Jayce and EnchantressOfNumbers like this.
  13. I got a video of a Gm recorded at last so why not
    upload_2016-9-18_22-6-25.png

    I managed level stops extremely well except for in one very obvious section. I'm actually quite proud of that lmao; 2 seconds of level stop from 500-999 is really surprising to me.

    Decent number of singles. The number of singles from 500-999 was just under double that of 0-500, so I should work on that. The section with the highest number of singles being 500-600 is no surprise considering you have to get the stack in some form of 20G friendliness. Some bad decisions in 400-500 and 600-700 led to a large number, and the tail of some misdrops in 800-900 hurt the 900-999 section.
    I dare say I managed a lot of situations concerning singles very well in converting them into doubles. Not really much else to say there.
    Very low number of triples. The number crept up in the latter half which I'll blame my love for triple rotations for.
    Nice number of Tetrises, though not enough for a would-be gold SK, sadly. I'm happy I managed to get at least one in every section. I find the large number of Tetrises in 100-200 amusing for some reason.

    No section times under a minute, but I kept fairly consistent with them from 500 onwards. 200-300 is the outsider due again to the level stop.

    All in all I'm quite happy with this run. I can easily bring it under 11:30 without much hassle, though I can see the grind for sub-11 being tough. The majority of time can be saved through clearing more Tetrises, less singles, and getting rid of that one level stop. I'm also quite proud of that score.

     
  14. kai

    kai

    EoN wouldn't let me get away with postponing this any further, so here goes! ^^

    I took the timings by stepping through the youtube video and I wouldn't bet my life that I didn't mess up any of the spreadsheet formulas but this should be roughly correct!

    Some sections had a gross amount of singles. If I get rid of a lot of those and just don't get level stopped for 18(!) whole seconds and then some, the sub11 should be free really!

    I guess I better look up that "avoiding the single" thread ^^

    [​IMG]
     
    EnchantressOfNumbers likes this.
  15. [​IMG]
    Here's one for my latest PB. Pretty consistent pace except for the one section (800-900) where I screwed up massively and nearly died. I think this shows that I should be able to get just under 9:40 with my current skill.
     
  16. Here's an analysis of my 9:42:56, with a lot more data collected about pieces placed and stack height.



     
  17. I finally did some line clear analysis (mostly for JBrom's singles challenge) of my new record. (See if you can spot the math guy among the computer scientists!)

    Even though my 20G time is still shorter than my lowG time, I expect much of that is simply due to drop time. It looks like I still have plenty of room to minimize singles in 20G. Given my 20G mode record is practically tied with my standard TGM1 time, I imagine the difference was sloppier play with more singles, enough to match the extra placement time in lowG. (Line clear delays were negligible in this run.) Moral of the story: It's time to start focusing on fixing mistakes via triples & Tetrises when at all possible.

    tgm analysis.jpg


    Here's the run that goes with it:

     
    jack13berry and Qlex like this.

Share This Page