I just purchased a 3 day trial and after a few hours of testing, it's looking very positive in comparison with the old version of Hearthbuddy. So far I have 2 suggestions. Hoovering over cards might be a little bit extended. What I mean is, currently when it's your turn, the bot hoovers only on the card it is going to pick, when normal players tend to (at least the majority) hoover over cards randomly, while deciding which one to play. moving back and forth with the mouse over the entire deck. This might be seen as a pattern. Second thing is the speed. I know it's been probably discussed, but the scenario where you have a bunch of cards on the board and when you start attacking with them, with few seconds in between each attack, is so obvious for trained eye and I spotted 2, or 3 players doing it and was thinking they were botting. Now I know, they're using HB The AI is far from good. A few missplays every game and sometimes moves so stupid, not even a bad player would make, also might bring some attention. At the end I'd mention emotes, but I know it's being worked on, so GL. Like I said earlier, the bot is definitely better, but still needs to be polished. To all the people more concerned about their accounts and afraid of botting, don't use HB yet. This little things I mentioned and probably a few more I didn't catch might trigger a report and ban IMO. What's your opinnions? P.S It would be nice to have a thread for suggestions.
Excellent points, kaime. I'd agree with your ending statement about not using Hearthbuddy yet. I took an extended break after the GvG downtime in order to see how the bot initially fared, and then I began botting again late January. I'll second two key points you made with regards to the bots' appearance: it needs to hover over cards longer and it needs to attack both a) more quickly and b) a bit more randomly. Every attack should not be the same amount of time. Furthermore, and I do realize that this is likely difficult to achieve, but if the bot is going to actually resemble a human playing then it should occasionally hover over one card then switch to another. It also should target something then target something else. It needs to look like it's actually thinking and attempting to decide whether or not a particular move is worth it. As it stands, if I encountered someone using HB I would likely be able to discern that it's a bot -- even if I hadn't used HB previously. At the very least I'd be extremely suspicious. Players know full well that bots can auto-greet and target, too. Run it at your own risk. Regarding the AI, I have some feedback about specific cards that I was gathering while I was grinding out Golden Druid wins. I'll post it here just in case it's helpful. Even if the developers are already aware of the issues, it's good information for players to know when attempting to build druid decks. Druid in general is a very difficult class to bot effectively because of the myriad of decision trees its cards have. Overall the bot does an admirable job in making decisions, but as expected it's not entirely perfect. I had a very difficult time building a deck for Druid that had an acceptable win rate. As a side note, I don't auto-concede (a bit worried about detection), so my win-rate also suffered due to the fact that I was consistently around rank 9 or 10 for the entire grind. Druid Cards Innervate Standard card in most druid decks. The problem with this card is that 75% of the time the bot will use it in a semi-coherent fashion, but 25% of the time it'll make really really terrible decisions with Innervate. Because I couldn't guarantee good decisions, I eventually stopped running this card in my deck. There were countless times when it used Innervate + Hero Power mid-game when it should have saved it for future runs. Wild Growth The fulcrum of Ramp/Force Druid variants, the bot doesn't understand this card well enough to get good use out of it. It doesn't mulligan aggressively for it, and it won't weave it in properly during the early and mid-game in order to give you full advantage of the card. Also will play it at very inopportune times, like on turn 9 when it would have been far better to just save it for turn10+ card draw. Force of Nature Standard card in most druid decks these days -- Force Druid is one of the few viable Druid options left right now in the current meta. The bot, however, cannot play a Force Druid properly, and as a result this card is essentially a wasted slot in your deck. It doesn't understand how critical it is to combo it with Savage Roar, and so it'll often just use it by itself recklessly. It's quick to use it defensively, which isn't generally good, and it also will use it offensively without having lethal damage (and no follow combo for the next turn). Nourish Another really puzzling card due to how the bot uses it. Most of the time it's fine and will use it for card draw...however, I witnessed some obscenely bad blunders with the card and was forced to not use it. The highlight of its Nourish blunders was using it for 2 extra mana crystals on turn 10. If my opponent didn't think I was botting before, it was all but confirmed with that hideous play. @_@ Savage Roar The bot doesn't understand how to use this efficiently. If it's in the hand and it has the mana, it will use it randomly in the mid and late game to influence favorable board trades. This isn't always bad, but the point of Savage Roar is securing lethal, not gaining small amounts of extra board control. Another really problematic issue that HB has with Savage Roar is that if Savage Roar is in the opening hand it will never mulligan it. Perhaps it's just checking the mana cost and considering it a good choice for the mid-game, I'm not certain. But at no point should a Druid keep Savage Roar in their opening hand. Cenarius Overall the bot's play with Cenarius is acceptable -- to the point at which I kept Cenarius in my final deck that got me golden Druid. That being said, there are some oddities in which the bot does not assess the board situation before making choices with Cenarius' power. The worst play I've seen was when the bot played Cenarius on a completely empty board and used Demigod's Favor. That's really bad, because it's another very obvious admission of guilt. Ancient of Lore The bot doesn't necessarily use Ancient of Lore badly as much as it tends to over prioritize it and its card draw. It will tend to favor the card draw of the Lore over almost any other play, regardless as to how many cards it has in its hand. This can be really troublesome when you've got much better, higher pressure lategame cards that could be played on the same turn. The bot's threshold for using the heal on the Ancient of Lore is also obscenely low, and as a result I would consistently die to Hunters and Zoolocks due to the fact that the bot considered ~12 health "good enough" and chose card draw over the heal. Ancient of Lore is still worth running in a Druid deck, but just be aware of these issues. Misc. Cards Bloodmage Thalnos The bot doesn't understand this card at all. Whenever I included Thalnos in one of my decks the bot would play Thalnos at the end of its turn, after it had already cast its spells. I think it goes without saying that this is really, really bad. Leeroy Jenkins The bot will use this card if it doesn't have a better play -- regardless as to whether or not it has lethal. After witnessing a turn 5 Leeroy play against an opponent that was above 20 health, I quickly dropped this card from any future HB decks. Mechwarper The bot is really close to understanding this card perfectly, but it still occasionally will coin out a different two drop (Annoy-o-tron, for instance) when Mechwarper was the obvious choice. That's all the data I've gathered so far. I'm taking another break from HB while it gets more fine-tuned. I think it's still a bit too risky to use regularly. Hopefully the information here is taken as something meant to help HB's development and isn't seen as some form of complaining. I appreciate the hard work being done on HB and think it's got a fantastic foundation on which to improve. For all of the poor AI plays mentioned regarding the previous cards, I've also seen some fantastic and clever moves come out from the bot. The bot is very impressive overall. Just needs some more time to get fully fleshed out. Best of luck to the developers, and thanks for all the hard work!
For individual cards, it would be the best to attach a log of a missplay in Silverfish thread I think. Well I am dropping the bot for now. Will check a trial again in a month or so
