add LayerNorm layer

config.lua

@@ -12,48 +12,62 @@ local function intmap(x)
     return math.pow(10, x / 2)
 end
 
-local cfg = {
-    log_fn = 'log.csv', -- can be nil to disable logging.
-
+local common_cfg = {
     defer_prints = true,
 
     playable_mode = false,
-    start_big = false, --true
-    starting_lives = 0, --1
+    playback_mode = false,
+    start_big = false,
+    starting_lives = 0,
 
-    init_zeros = true, -- instead of he_normal noise or whatever.
     frameskip = 4,
     -- true greedy epsilon has both deterministic and det_epsilon set.
     deterministic = false, -- use argmax on outputs instead of random sampling.
     det_epsilon = false, -- take random actions with probability eps.
+    layernorm = false,
 
+    init_zeros = true, -- instead of he_normal noise or whatever.
     graycode = false,
-    epoch_trials = 64 * (7/8),
-    epoch_top_trials = 40 * (7/8), -- new with ARS.
     unperturbed_trial = true, -- do a trial without any noise.
     negate_trials = true, -- try pairs of normal and negated noise directions.
-    time_inputs = true, -- binary inputs of global frame count
     -- ^ note that this now doubles the effective trials.
-    deviation = intmap(-3),
-    learning_rate = intmap(-4),
-    weight_decay = intmap(-6),
+    time_inputs = true, -- binary inputs of global frame count
 
-    adamant = true,  -- run steps through AMSgrad.
-    adam_b1 = math.pow(10, -1 / 15),
-    adam_b2 = math.pow(10, -1 / 100),
-    adam_eps = intmap(-8),
-    adam_debias = false,
+    adamant = false, -- run steps through AMSgrad.
 
-    cap_time = 222, --400
+    cap_time = 300,
     timer_loser = 1/2,
-    decrement_reward = false, -- bad idea, encourages mario to kill himself
+    decrement_reward = false, -- bad idea, encourages mario to run into goombas.
+}
 
-    playback_mode = false,
+local cfg = {
+    log_fn = 'log.csv', -- can be nil to disable logging.
+    params_fn = nil, -- can be nil to generate based on param count.
+
+    deterministic = true,
+
+    epoch_trials = 20,
+    epoch_top_trials = 10,
+    learning_rate = 1.0,
+
+    deviation = 0.1,
+    weight_decay = 0.004,
+
+    cap_time = 300,
+    starting_lives = 1,
 }
 
 -- TODO: so, uhh..
 -- what happens when playback_mode is true but unperturbed_trial is false?
 
+setmetatable(cfg, {
+    __index = function(t, n)
+        if common_cfg[n] ~= nil then return common_cfg[n] end
+        if n == 'params_fn' then return nil end
+        error("cannot use undeclared config '" .. tostring(n) .. "'", 2)
+    end
+})
+
 cfg.epoch_top_trials = math.min(cfg.epoch_trials, cfg.epoch_top_trials)
 
 cfg.eps_start = 1.0 * cfg.frameskip / 64
@@ -62,8 +76,7 @@ cfg.eps_frames = 1000000
 cfg.enable_overlay = cfg.playable_mode
 cfg.enable_network = not cfg.playable_mode
 
-return setmetatable(cfg, {
-    __index = function(t, n)
-        error("cannot use undeclared config '" .. tostring(n) .. "'", 2)
-    end
-})
+assert(not cfg.ars_lips or cfg.unperturbed_trial,
+       "cfg.unperturbed_trial must be true to use cfg.ars_lips")
+
+return cfg

main.lua

@@ -227,6 +227,7 @@ local function make_network(input_size)
     nn_ty:feed(nn_y)
 
     nn_y = nn_y:feed(nn.Dense(128))
+    if cfg.layernorm then nn_y = nn_y:feed(nn.LayerNorm()) end
     if cfg.deterministic then
         nn_y = nn_y:feed(nn.Relu())
     else

nn.lua

@@ -311,6 +311,7 @@ local Gelu = Layer:extend()
 local Dense = Layer:extend()
 local Softmax = Layer:extend()
 local Embed = Layer:extend()
+local LayerNorm = Layer:extend()
 
 function Weights:init(weight_init)
     self.weight_init = weight_init
@@ -618,6 +619,42 @@ function Embed:forward(X)
     return Y
 end
 
+function LayerNorm:init(eps)
+    Layer.init(self, "LayerNorm")
+    if eps == nil then eps = 1e-5 end
+    assert(type(eps) == "number")
+    self.eps = eps
+end
+
+function LayerNorm:reset_cache(bs)
+    self.bs = bs
+    self.cache = cache(bs, self.shape_out)
+end
+
+function LayerNorm:forward(X)
+    local bs = checkshape(X, self.shape_in)
+    if self.bs ~= bs then self:reset_cache(bs) end
+
+    local mean = 0
+    for i, v in ipairs(X) do
+        mean = mean + v / #X
+    end
+
+    local var = 0
+    for i, v in ipairs(X) do
+        local delta = v - mean
+        self.cache[i] = delta
+        var = var + delta * delta / #X
+    end
+
+    local std = sqrt(var + self.eps)
+    for i, v in ipairs(self.cache) do
+        self.cache[i] = v / std
+    end
+
+    return self.cache
+end
+
 function Model:init(nodes_in, nodes_out)
     assert(#nodes_in > 0, #nodes_in)
     assert(#nodes_out > 0, #nodes_out)
@@ -764,14 +801,15 @@ return {
     traverse = traverse,
     traverse_all = traverse_all,
 
-    Weights = Weights,
-    Layer   = Layer,
-    Model   = Model,
-    Input   = Input,
-    Merge   = Merge,
-    Relu    = Relu,
-    Gelu    = Gelu,
-    Dense   = Dense,
-    Softmax = Softmax,
-    Embed   = Embed,
+    Weights     = Weights,
+    Layer       = Layer,
+    Model       = Model,
+    Input       = Input,
+    Merge       = Merge,
+    Relu        = Relu,
+    Gelu        = Gelu,
+    Dense       = Dense,
+    Softmax     = Softmax,
+    Embed       = Embed,
+    LayerNorm   = LayerNorm,
 }