add mean adaptation hyperparameter

config.lua

@@ -32,6 +32,9 @@ local common_cfg = {
     time_inputs = true, -- binary inputs of global frame count
     normalize_inputs = false,
 
+    learning_rate = 1.0,
+    mean_adapt = 1.0, -- for xNES
+
     es = 'ars',
     ars_lips = false,
     adamant = false, -- run steps through AMSgrad.

main.lua

@@ -456,6 +456,8 @@ local function init()
         -- maybe there'll be a patch for FCEUX in the future.
         es = xnes.Xnes(network.n_param, cfg.epoch_trials,
                        cfg.learning_rate, cfg.deviation, cfg.negate_trials)
+        -- TODO: clean this up into an interface:
+        es.mean_adapt = cfg.mean_adapt
     elseif cfg.es == 'ars' then
         es = ars.Ars(network.n_param, cfg.epoch_trials, cfg.epoch_top_trials,
                      cfg.learning_rate, cfg.deviation, cfg.negate_trials)

xnes.lua

@@ -71,17 +71,15 @@ function Xnes:init(dims, popsize, learning_rate, sigma, antithetic)
     --self.log_sigma = log(self.sigma)
     --self.log_covars = zeros{dims, dims}
     --for i, v in ipairs(self.covars) do self.log_covars[i] = log(v) end
+
+    self.mean_adapt = 1.0
 end
 
-function Xnes:params(new_mean, new_covars)
+function Xnes:params(new_mean)
     if new_mean ~= nil then
         assert(#self.mean == #new_mean, "new parameters have the wrong size")
         for i, v in ipairs(new_mean) do self.mean[i] = v end
     end
-    if new_covars ~= nil then
-        -- TODO: assert determinant of new_covars is 1.
-        error("TODO")
-    end
     return self.mean
 end
 
@@ -183,7 +181,7 @@ function Xnes:tell(scored, noise)
 
     local dotted = dot_mv(self.covars, g_delta)
     for i, v in ipairs(self.mean) do
-        self.mean[i] = v + self.sigma * dotted[i]
+        self.mean[i] = v + self.mean_adapt * self.sigma * dotted[i]
     end
 
     --[[