OK - continuing.
Now - what I typically pay attention to when I'm watching the vid or game that got me exciting to work on my skills with a certain champ, is how do they combo, what is their typical playstyle, do they have a strong engage, etc. I don't theory craft any of this initially - I just look for solid play and then seek to emulate that. It's kind of like when you see a fort pot riven. Well, you play fort pot riven a bit differently than you would a normal riven. For pot rivens will be hyper aggressive, typically running ignite, and be looking for any quick opening to kill you off quickly. So, I pay attention to that playstyle and I see how they accomplish that and when they go ape shite for the kill.
After I've done a little bit of studying, then its time to fire up a custom game against bots.
AI match setup
Now, you can't emulate a real game regarding laning/playstyle. Here are some tips, though. If you want to practice mid, you can load a full custom game with as a 5v5 with bots on all teams. Just head to mid and you will be 1v1 mid. If you want to practice top, load 3 AI's on the enemy team and 1 or 0 bots on your team. Your team's bot will head to bot lane. If you have any other number of bots, one bot will go top with you and you won't be 1v1. For practicing the jungle, well, you don't have any leash, so here's what I do. I'll load the game 5v5 with bots. Then immediately head to red/blue - wherever I'm going to start. If you are in a lane when the bots start moving, well, they might readjust and you could end up with no one covering mid. So, just be out of sight in the jungle and you'll have 1 ai go top, 1 ai go mid, and 2 go ai. That doesn't really matter much - just an FYI. Anyway, you don't have a leash, so you typically have to play your jungle a bit differently but can end with the same basic result provided you weren't going to do an immediately red/blue or blue/red gank. Basically, I just pick machete + red pots and go directly to blue or red. I'll smite immediately when it spawns and then cast a health pot or 2. Once you clear it, just proceed to the smaller jungle camps while you wait for your smite to come back up. Just work the jungle over. Then, when smite is up again, smite the remaining big buff and clear the camp. You should be in decent fighting shape at that point with good health/mana. Then, just continue your jungle route.
OK - so why bother playing against AI at all? Well, 2 reasons mostly. 1 - to familiarize yourself with the champ including what abilities you will get at what time, what you will shop for at what time, and, quite importantly in my mind, to get a feel for your damage output and what you can do. If you don't have a good understanding of your burst, you are going to suck if you try to all in. See, its often super obvious when you can pull off a kill - the enemy is low hp, flash it down, no barrier, w/e. You just go for the kill. But what about if you are top or mid lane and the guy you are laning against has 400 hp. Do you know for sure that you can kill them early game based on your potential dps? You probably don't unless you are very familiar with the character. Anyway, in short, the bot setup is great for figuring out your early game damage potential. That's something I'd always prefer to know going into a game instead of learning on the fly. It's also excellent for working on last hitting with a champ you aren't familiar with. Also something that is very good to know before playing a normal game.
Now, I'd think it goes without saying, but I'll say it anyway - bot matches don't really make you are better player. Oh great - I just got 42 kills in my bot match. This means both jack and shite. Bots are stupid and 100% predictable. If you play enough bot matches, you can easily pick out the patterns. Kill a tower and the AI will drop what its doing to swarm to that tower. Do this, ai does that. Racking up kills means nothing and those kills don't really translate over to normal games because people generally aren't that freaking stupid.
Anyway, to summarize this section, I think its a great idea to fire up a custom game against bots to try out new builds/characters or refresh yourself. You can get a good idea early game of your damage potential, can practice last hitting, and can practice combos, etc. Outside of that, the bot matches aren't all that useful for training imo.
Theory crafting
Well, this still fits a smidge in the AI match section to some degree, but its a lot more work, but is exceptional at improving your understanding of the game, items, abilities, and characters. The goal is to find the best possible build, ability order, runes/masteries, shop times, item purchase order, for a given champion. The complexity occurs when you try to do that in detail.
So, we'll start with a basic general premise and build from there. Good players likely have some idea on what they building and why. This isn't always the case, but you see certain setups become the standard. A 21/9/0 adc is likely the most commonly accepted adc setup. Most folks know this so they simply emulate it because that's what X player does and he's good. With further thought, you see why X player chooses that setup. You see the benefits gained from these points in the mastery tree over that point and come to agree with the setup (provided you thought things through this far).
Ok - so then you get into alternative adc setups. I've played and tested out 2 variant adc setups with mixed results. One is a setup that doublelift often runs on a vayne where you go 21/1/8 and get bonus cooldown reduction on your summoners. You trade off early game tankiness for increased odds of winning a fight after you've burned the enemy summoners as your summoners will be up more quickly. This setup is best if you are confident you can outplay a particular player. BUT - you maintain the same AD burst potential. It's a situational pick. Another setup I tried was X/X/X (can't remember, but went pretty deep into support maxing out the greed setup). The advantage to this particular setup was primarily for early/mid. You extra money would allow you 1-2 pots (can't remember) and potentially a biscuit. This = more sustain in lane and better recovery if you blow it. The bad tradeoff with this is that your damage output is lower late game as a result. For me, this was a decent setup at one point while I was weaker as an adc. It gave me enough staying power to safely farm and potentially lock down a kill. PLUS - added money for more quickly buying items.
Anyway, my latest theory crafting experiment was all about Galio. What peaked my interest was the horrible thing that brad was doing to me when I was goofing around with ap corki. He took galio and came up with a basic setup that pretty much completely countered my silly arse corki build. The concept is that galio's bulwark combined with spirit visage provides a stupidly ridiculous amount of healing, particularly if anyone uses dots like liandrys against him. With corki, that pretty much meant that every 4 seconds I'd be healing him past any possible level of burst. It was sick. Anyway, the underlying concept started with figuring out if galio with spirit visage would be practical in a normal game and under what circumstances. I won't cover all the details here, but through theory crafting, I figured out the optimal runes/masteries, the optimum ability order, and a good chunk of the best itemization for him, and general rules re: when to play pick him. All of that took many custom games and quite a bit of thought of where galio fits in a team comp. But, as a result, I know the champ pretty much inside and out and my skill level with him has improved tremendously.
Game Time
After all that (I don't always go so deep as to theory craft up a storm, mind you - its pretty time consuming), I fire up a normal and try out what I've learned/crafted. Based on the results, I might adjust this or that. Normal games are very useful for better understanding when you should buy what item. For instance, on galio, I learned that I would always rush chalice, but there is some flexibility immediately after that for a certain period of time. Anyway, that's about it. Thanks for reading.