smbot/main.lua

local globalize = require("strict")

-- configuration.

local cfg = require("config")
local gcfg = require("gameconfig")

-- state.

local epoch_i = 0
local base_params
local trial_i = -1 -- NOTE: trial 0 is an unperturbed trial, if enabled.
local trial_neg = true
local trial_noise = {}
local trial_rewards = {}
local trials_remaining = 0
local mom1 -- first moments in AMSgrad.
local mom2 -- second moments in AMSgrad.
local mom2max -- running element-wise maximum of mom2.

local trial_frames = 0
local total_frames = 0
local lagless_count = 0

local force_start = false
local force_start_old = false

local startsave = savestate.create(1)

local poketime = false
local max_time

local jp

local screen_scroll_delta
local reward
--local all_rewards = {}

local powerup_old
local status_old
local coins_old
local score_old

local once = false
local reset = true

local state_old = ''
local last_trial_state

-- localize some stuff.

local assert = assert
local print = print
local ipairs = ipairs
local pairs = pairs
local select = select
local open = io.open
local abs = math.abs
local floor = math.floor
local ceil = math.ceil
local min = math.min
local max = math.max
local exp = math.exp
local pow = math.pow
local log = math.log
local sqrt = math.sqrt
local random = math.random
local randomseed = math.randomseed
local insert = table.insert
local remove = table.remove
local unpack = table.unpack or unpack
local sort = table.sort

local band = bit.band
local bor = bit.bor
local bxor = bit.bxor
local bnot = bit.bnot
local lshift = bit.lshift
local rshift = bit.rshift
local arshift = bit.arshift
local rol = bit.rol
local ror = bit.ror

local emu = emu
local gui = gui

local util = require("util")
local argmax           = util.argmax
local argsort          = util.argsort
local calc_mean_dev    = util.calc_mean_dev
local clamp            = util.clamp
local copy             = util.copy
local empty            = util.empty
local lerp             = util.lerp
local softchoice       = util.softchoice
local unperturbed_rank = util.unperturbed_rank

local game = require("smb")
game.overlay = cfg.enable_overlay

-- utilities.

local log_map = {
    epoch = 1,
    trial_mean = 2,
    trial_std = 3,
    delta_mean = 4,
    delta_std = 5,
    step_std = 6,
    adam_std = 7,
    weight_mean = 8,
    weight_std = 9,
    test_trial = 10,
}

local function log_csv(t)
    if cfg.log_fn == nil then return end
    local f = open(cfg.log_fn, 'a')
    if f == nil then error("Failed to open log file "..cfg.log_fn) end
    local values = {}
    for k, v in pairs(t) do
        local i = log_map[k]
        if i == nil then error("Unexpected log key "..tostring(k)) end
        values[i] = v
    end
    for k, i in pairs(log_map) do
        if values[i] == nil then error("Missing log key "..tostring(k)) end
    end
    for i, v in ipairs(values) do
        f:write(tostring(v))
        if i ~= #values then f:write(",") end
    end
    f:write('\n')
    f:close()
end

-- network parameters.

package.loaded['nn'] = nil -- DEBUG
local nn = require("nn")

local network
local nn_x, nn_tx, nn_ty, nn_y, nn_z
local function make_network(input_size)
    nn_x = nn.Input({input_size})
    nn_tx = nn.Input({gcfg.tile_count})
    nn_ty = nn_tx:feed(nn.Embed(#game.valid_tiles, 2))
    nn_y = nn.Merge()
    nn_x:feed(nn_y)
    nn_ty:feed(nn_y)

    nn_y = nn_y:feed(nn.Dense(128))
    if cfg.deterministic then
        nn_y = nn_y:feed(nn.Relu())
    else
        nn_y = nn_y:feed(nn.Gelu())
    end
    if cfg.layernorm then nn_y = nn_y:feed(nn.LayerNorm()) end

    nn_z = nn_y
    nn_z = nn_z:feed(nn.Dense(#gcfg.jp_lut))
    nn_z = nn_z:feed(nn.Softmax())
    return nn.Model({nn_x, nn_tx}, {nn_z})
end

-- learning and evaluation.

local function prepare_epoch()
    base_params = network:collect()
    if cfg.playback_mode then return end

    print('preparing epoch '..tostring(epoch_i)..'.')
    empty(trial_noise)
    empty(trial_rewards)

    local precision = (pow(cfg.deviation, 1/-0.51175585) - 8.68297257) / 1.66484392
    if cfg.graycode then
        print(("chosen precision: %.2f"):format(precision))
    end

    for i = 1, cfg.epoch_trials do
        local noise = nn.zeros(#base_params)
        if cfg.graycode then
            for j = 1, #base_params do
                noise[j] = exp(-precision * nn.uniform())
            end
            for j = 1, #base_params do
                noise[j] = nn.uniform() < 0.5 and noise[j] or -noise[j]
            end
        else
            for j = 1, #base_params do
                noise[j] = cfg.deviation * nn.normal()
            end
        end
        trial_noise[i] = noise
    end
    trial_i = -1
end

local function load_next_pair()
    trial_i = trial_i + 1
    if trial_i == 0 and not cfg.unperturbed_trial then
        trial_i = 1
        trial_neg = true
    end

    local W = copy(base_params)

    if trial_i > 0 then
        if trial_neg then
            local noise = trial_noise[trial_i]
            for i, v in ipairs(base_params) do
                W[i] = v + noise[i]
            end

        else
            trial_i = trial_i - 1
            local noise = trial_noise[trial_i]
            for i, v in ipairs(base_params) do
                W[i] = v - noise[i]
            end
        end

        trial_neg = not trial_neg
    end

    network:distribute(W)
end

local function load_next_trial()
    if cfg.negate_trials then return load_next_pair() end
    trial_i = trial_i + 1
    local W = copy(base_params)
    if trial_i == 0 and not cfg.unperturbed_trial then
        trial_i = 1
    end
    if trial_i > 0 then
        print('loading trial', trial_i)
        local noise = trial_noise[trial_i]
        for i, v in ipairs(base_params) do
            W[i] = v + noise[i]
        end
    else
        print("test trial")
    end
    network:distribute(W)
end

local function collect_best_indices()
    -- select one (the best) reward of each pos/neg pair.
    local best_rewards = {}
    if cfg.negate_trials then
        for i = 1, cfg.epoch_trials do
            local ind = (i - 1) * 2 + 1
            local pos = trial_rewards[ind + 0]
            local neg = trial_rewards[ind + 1]
            best_rewards[i] = max(pos, neg)
        end
    else
        best_rewards = copy(trial_rewards)
    end

    local indices = argsort(best_rewards, function(a, b) return a > b end)

    for i = cfg.epoch_top_trials + 1, #best_rewards do indices[i] = nil end
    return indices
end

local function kinda_lipschitz(dir, pos, neg, mid)
    local _, dev = calc_mean_dev(dir)
    local c0 = neg - mid
    local c1 = pos - mid
    local l0 = abs(3 * c1 + c0)
    local l1 = abs(c1 + 3 * c0)
    return max(l0, l1) / (2 * dev)
end

local function make_step_paired(rewards, current_cost)
    local step = nn.zeros(#base_params)
    local _, reward_dev = calc_mean_dev(rewards)
    if reward_dev == 0 then reward_dev = 1 end

    for i = 1, cfg.epoch_trials do
        local ind = (i - 1) * 2 + 1
        local pos = rewards[ind + 0]
        local neg = rewards[ind + 1]
        local reward = pos - neg
        if reward ~= 0 then
            local noise = trial_noise[i]

            if cfg.ars_lips then
                local lips = kinda_lipschitz(noise, pos, neg, current_cost)
                reward = reward / lips / cfg.deviation
            else
                reward = reward / reward_dev
            end

            for j, v in ipairs(noise) do
                step[j] = step[j] + reward * v / cfg.epoch_top_trials
            end
        end
    end
    return step
end

local function make_step(rewards)
    local step = nn.zeros(#base_params)
    local _, reward_dev = calc_mean_dev(rewards)
    if reward_dev == 0 then reward_dev = 1 end

    for i = 1, cfg.epoch_trials do
        local reward = rewards[i] / reward_dev
        if reward ~= 0 then
            local noise = trial_noise[i]

            for j, v in ipairs(noise) do
                step[j] = step[j] + reward * v / cfg.epoch_top_trials
            end
        end
    end
    return step
end

local function amsgrad(step) -- in-place!
    if mom1 == nil then mom1 = nn.zeros(#step) end
    if mom2 == nil then mom2 = nn.zeros(#step) end
    if mom2max == nil then mom2max = nn.zeros(#step) end

    local b1_t = pow(cfg.adam_b1, epoch_i)
    local b2_t = pow(cfg.adam_b2, epoch_i)

    -- NOTE: with LuaJIT, splitting this loop would
    --       almost certainly be faster.
    for i, v in ipairs(step) do
        mom1[i] = cfg.adam_b1 * mom1[i] + (1 - cfg.adam_b1) * v
        mom2[i] = cfg.adam_b2 * mom2[i] + (1 - cfg.adam_b2) * v * v
        mom2max[i] = max(mom2[i], mom2max[i])
        if cfg.adam_debias then
            local num = (mom1[i] / (1 - b1_t))
            local den = sqrt(mom2max[i] / (1 - b2_t)) + cfg.adam_eps
            step[i] = num / den
        else
            step[i] = mom1[i] / (sqrt(mom2max[i]) + cfg.adam_eps)
        end
    end
end

local function learn_from_epoch()
    print()
    --print('rewards:', trial_rewards)

    --for _, v in ipairs(trial_rewards) do insert(all_rewards, v) end

    local current_cost = trial_rewards[0] -- may be nil!

    if cfg.unperturbed_trial then
        local nth_place = unperturbed_rank(trial_rewards, current_cost)

        -- a rank of 1 means our gradient is uninformative.
        print(("test trial: %d out of %d"):format(nth_place, #trial_rewards))
    end

    local delta_rewards = {} -- only used for logging.
    if cfg.negate_trials then
        for i = 1, cfg.epoch_trials do
            local ind = (i - 1) * 2 + 1
            local pos = trial_rewards[ind + 0]
            local neg = trial_rewards[ind + 1]
            delta_rewards[i] = abs(pos - neg)
        end
    end

    local indices = collect_best_indices()
    print("best trials:", indices)

    local top_rewards = {}
    for i = 1, #trial_rewards do top_rewards[i] = 0 end
    for _, ind in ipairs(indices) do
        local sind = (ind - 1) * 2 + 1
        top_rewards[sind + 0] = trial_rewards[sind + 0]
        top_rewards[sind + 1] = trial_rewards[sind + 1]
    end
    --print("top:", top_rewards)

    if cfg.negate_trials then
        local top_delta_rewards = {} -- only used for printing.
        for i, ind in ipairs(indices) do
            local sind = (ind - 1) * 2 + 1
            top_delta_rewards[i] = abs(top_rewards[sind + 0] - top_rewards[sind + 1])
        end
        print("best deltas:", top_delta_rewards)
    end

    local step
    if cfg.negate_trials then
        step = make_step_paired(top_rewards, current_cost)
    else
        step = make_step(top_rewards)
    end

    local step_mean, step_dev = calc_mean_dev(step)
    print("step mean:", step_mean)
    print("step stddev:", step_dev)

    local momstep_mean, momstep_dev = 0, 0
    if cfg.adamant then
        amsgrad(step)
        momstep_mean, momstep_dev = calc_mean_dev(step)
        print("amsgrad mean:", momstep_mean)
        print("amsgrad stddev:", momstep_dev)
    end

    for i, v in ipairs(base_params) do
        base_params[i] = v + cfg.learning_rate * step[i] - cfg.weight_decay * v
    end

    local trial_mean, trial_std = calc_mean_dev(trial_rewards)
    local delta_mean, delta_std = calc_mean_dev(delta_rewards)
    local weight_mean, weight_std = calc_mean_dev(base_params)

    log_csv{
        epoch = epoch_i,
        trial_mean = trial_mean,
        trial_std = trial_std,
        delta_mean = delta_mean,
        delta_std = delta_std,
        step_std = step_dev,
        adam_std = momstep_dev,
        weight_mean = weight_mean,
        weight_std = weight_std,
        test_trial = current_cost or 0,
    }

    if cfg.enable_network then
        network:distribute(base_params)
        network:save(cfg.params_fn)
    else
        print("note: not updating weights in playable mode.")
    end

    print()
end

local function joypad_mash(button)
    local jp_mash = {
        up = false,
        down = false,
        left = false,
        right = false,
        A = false,
        B = false,
        select = false,
        start = false,
    }
    assert(jp_mash[button] == false, "invalid button: "..tostring(button), 1)
    jp_mash[button] = emu.framecount() % 2 == 1
    joypad.write(1, jp_mash)
end

local function do_reset()
    local state = game.get_state()
    -- be a little more descriptive.
    if state == 'dead' and game.get_timer() == 0 then state = 'timeup' end

    if trial_i >= 0 then
        if trial_i == 0 then
            print('test trial reward:', reward, "("..state..")")
        elseif cfg.negate_trials then
            --local dir = trial_neg and "negative" or "positive"
            --print('trial', trial_i, dir, 'reward:', reward, "("..state..")")

            if trial_neg then
                local pos = trial_rewards[#trial_rewards]
                local neg = reward
                local fmt = "trial %i rewards: %+i, %+i (%s, %s)"
                print(fmt:format(trial_i, pos, neg, last_trial_state, state))
            end
            last_trial_state = state
        else
            print('trial', trial_i, 'reward:', reward, "("..state..")")
        end
    else
        print("reward:", reward, "("..state..")")
    end

    if trial_i >= 0 then
        if trial_i == 0 or not cfg.negate_trials then
            trial_rewards[trial_i] = reward
        else
            trial_rewards[#trial_rewards + 1] = reward
        end
    end

    if epoch_i == 0 or (trial_i == cfg.epoch_trials and trial_neg) then
        if epoch_i > 0 then learn_from_epoch() end
        if not cfg.playback_mode then epoch_i = epoch_i + 1 end
        prepare_epoch()
    end

    if game.get_state() == 'loading' then game.advance() end -- kind of a hack.
    reward = 0
    powerup_old = game.R(0x754)
    status_old = game.R(0x756)
    coins_old = game.R(0x7ED) * 10 + game.R(0x7EE)
    score_old = game.get_score()

    -- set number of lives. (mario gets n+1 chances)
    game.W(0x75A, cfg.starting_lives)

    if cfg.start_big then
        -- make mario "super".
        game.W(0x754, 0)
        game.W(0x756, 1)
    end

    max_time = min(6 * sqrt(480 / cfg.epoch_trials * (epoch_i - 1)) + 60, cfg.cap_time)
    max_time = ceil(max_time)

    if once then
        savestate.load(startsave)
    else
        savestate.save(startsave)
    end
    once = true

    jp = nil
    screen_scroll_delta = 0
    trial_frames = 0
    emu.frameadvance() -- prevents emulator from quirking up.

    load_next_trial()

    reset = false
end

local function init()
    network = make_network(gcfg.input_size)
    network:reset()
    network:print()
    print("parameters:", network.n_param)

    if cfg.init_zeros then
        local W = network:collect()
        for i, w in ipairs(W) do W[i] = 0 end
        network:distribute(W)
    end

    emu.poweron()
    emu.unpause()
    if not cfg.playable_mode then emu.speedmode("turbo") end

    while emu.framecount() < 195 do  -- FIXME: don't hardcode this.
        joypad_mash('start')
        emu.frameadvance()
    end
    --print(emu.framecount())

    local res, err = pcall(network.load, network, cfg.params_fn)
    if res == false then print(err) end
end

local function prepare_reset()
    if cfg.playback_mode then return end
    reset = true
end

local function draw_framecount(x, y, frames)
    local tf0 = frames % 1000
    local tf1 = (frames % 1000000 - tf0) / 1000
    local tf2 = (frames - tf0 - tf1) / 1000000
    gui.text(x, y, ("%03i,%03i,%03i"):format(tf2,tf1,tf0), '#FFFFFF', '#0000003F')
end

local function doit(dummy)
    local ingame_paused = game.get_state() == "paused"

    -- every few frames mario stands still, forcibly decrease the timer.
    -- this includes having the game paused.
    -- TODO: more robust. doesn't detect moonwalking against a wall.
    -- well, that shouldn't happen anymore now that i've disabled left+right.
    local timer = game.get_timer()
    if ingame_paused or random() > 1 - cfg.timer_loser and game.R(0x1D) == 0 and game.R(0x57) == 0 then
        timer = timer - 1
    end
    if not cfg.playback_mode then
        timer = clamp(timer, 0, max_time)
        if cfg.enable_network then
            game.set_timer(timer)
        end
    end

    draw_framecount(12, 215, total_frames)

    screen_scroll_delta = screen_scroll_delta + game.R(0x775)

    if dummy == true then
        -- don't invoke AI this frame. (keep holding the old inputs)
        gui.text(89, 16, ("%+5i"):format(reward), '#FFFFFF', '#0000003F')
        return
    end

    empty(game.sprite_input)
    empty(game.tile_input)
    empty(game.extra_input)

    -- TODO: check if mario is in a playable state.
    local x, y = game.getxy(0, 0x86, 0xCE, 0x6D, 0xB5)
    local powerup = game.R(0x754)
    local status = game.R(0x756)
    game.mark_sprite(x + 8, y + 24, -powerup - 1)

    local vx, vy = game.S(0x57), game.S(0x9F)
    insert(game.extra_input, vx * 16)
    insert(game.extra_input, vy * 16)

    if cfg.time_inputs then
        for i=2,5 do
            local v = band(trial_frames, lshift(1, i)) == 0 and -181 or 181
            insert(game.extra_input, v)
        end
    end

    game.handle_enemies()
    game.handle_fireballs()
    --game.handle_blocks() -- blocks being hit. not interactable; we don't care!
    game.handle_hammers()
    game.handle_misc()
    game.handle_tiles()

    local coins = game.R(0x7ED) * 10 + game.R(0x7EE)
    local coins_delta = coins - coins_old
    -- handle wrap-around.
    if coins_delta < 0 then coins_delta = 100 + coins - coins_old end
    -- remember that 0 is big mario and 1 is small mario.
    local powerup_delta = powerup_old - powerup
    -- 2 is fire mario.
    local status_delta = clamp(status - status_old, -1, 1)
    local flagpole_bonus = game.R(0xE) == 4 and cfg.frameskip or 0
    --local reward_delta = screen_scroll_delta + status_delta * 256 + flagpole_bonus
    local score_delta = game.get_score() - score_old
    if score_delta < 0 then score_delta = 0 end
    local reward_delta = screen_scroll_delta + score_delta + flagpole_bonus
    screen_scroll_delta = 0

    if cfg.decrement_reward and reward_delta == 0 then reward_delta = reward_delta - 1 end

    if not ingame_paused and game.get_state() ~= 'win_walking' then
        -- note that we exclude points gained walking into the castle.
        -- this way, we avoid adding the timer-based fireworks to our reward,
        -- which are basically unwanted noise due to the way they trigger.
        reward = reward + reward_delta
    end

    --gui.text(72, 12, ("%+4i"):format(reward_delta), '#FFFFFF', '#0000003F')
    gui.text(89, 16, ("%+5i"):format(reward), '#FFFFFF', '#0000003F')

    if game.get_state() == 'dead' and state_old ~= 'dead' then
        --print("dead. lives remaining:", game.R(0x75A, 0))
        if game.R(0x75A, 0) == 0 then prepare_reset() end
    end
    if game.get_state() == 'lose' then
        -- this shouldn't happen if we catch the deaths as above.
        print("ran out of lives.")
        if not cfg.playback_mode then prepare_reset() end
    end

    -- lose a point for every frame paused.
    --if ingame_paused then reward = reward - 1 end
    if ingame_paused then reward = reward - 402; prepare_reset() end

    -- if we've run out of time while the game is paused...
    -- that's cheating! unpause.
    force_start = ingame_paused and timer == 0

    local X = {}
    for i, v in ipairs(game.sprite_input) do insert(X, v / 256) end
    for i, v in ipairs(game.extra_input) do insert(X, v / 256) end
    nn.reshape(X, 1, gcfg.input_size)
    nn.reshape(game.tile_input, 1, gcfg.tile_count)

    trial_frames = trial_frames + cfg.frameskip
    if cfg.enable_network and game.get_state() == 'playing' or ingame_paused then
        total_frames = total_frames + cfg.frameskip

        local outputs = network:forward({[nn_x]=X, [nn_tx]=game.tile_input})

        local eps = lerp(cfg.eps_start, cfg.eps_stop, total_frames / cfg.eps_frames)
        if cfg.det_epsilon and random() < eps then
            local i = floor(random() * #gcfg.jp_lut) + 1
            jp = copy(gcfg.jp_lut[i], jp)
        else
            local choose = cfg.deterministic and argmax or softchoice
            local ind = choose(unpack(outputs[nn_z]))
            jp = copy(gcfg.jp_lut[ind], jp)
        end

        if force_start then
            jp = {
                up      = false,
                down    = false,
                left    = false,
                right   = false,
                A       = false,
                B       = false,
                start   = force_start_old,
                select  = false,
            }
        end
    end

    coins_old = coins
    powerup_old = powerup
    status_old = status
    force_start_old = force_start
    state_old = game.get_state()
    score_old = game.get_score()
end

init()

while true do
    gui.text(4, 12, game.get_state(), '#FFFFFF', '#0000003F')

    while gcfg.bad_states[game.get_state()] do
        -- mash the start button until we have control.
        joypad_mash('start')
        prepare_reset()

        game.advance()
        gui.text(4, 12, game.get_state(), '#FFFFFF', '#0000003F')

        while game.get_state() == "loading" do game.advance() end -- kind of a hack.
        state_old = game.get_state()
    end

    if reset then
        do_reset()
        lagless_count = 0
    end

    if not cfg.enable_network then
        -- infinite time cheat. super handy for testing.
        if game.R(0xE) == 8 then
            game.set_timer(667)
            poketime = true
        elseif poketime then
            poketime = false
            game.set_timer(1)
        end

        -- infinite lives.
        game.W(0x75A, 1)
    end

    local doot = jp == nil or lagless_count % cfg.frameskip == 0
    doit(not doot)

    -- jp might still be nil if we're not ingame or we're not playing.
    if jp ~= nil then joypad.write(1, jp) end

    game.advance() -- remember, this skips lag frames.
    lagless_count = lagless_count + 1
end