From bd7c52c927808953eb4234a701dea667eaed2c12 Mon Sep 17 00:00:00 2001
From: "E. Almqvist" <elalmqvist@gmail.com>
Date: Tue, 20 Oct 2020 15:54:34 +0200
Subject: [PATCH] Feed the AI random inputs per learn iteration

---
 rgbAI/lib/ailib/__pycache__/ai.cpython-38.pyc | Bin 4648 -> 4620 bytes
 rgbAI/lib/ailib/ai.py                         |  12 ++++++++----
 2 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/rgbAI/lib/ailib/__pycache__/ai.cpython-38.pyc b/rgbAI/lib/ailib/__pycache__/ai.cpython-38.pyc
index fa2344de7cd910272901498e9be0dfed05e42a35..20dfc6028f3a104ea33c3af99441dabdcdfbbe43 100644
GIT binary patch
delta 192
zcmZ3X(xbu`%FD~e00d88_r=SKZ{(ZB=f@7@Is<WW6p$!ks9|VktYs--tYK<qbYX~f
zsb#HU$zrNuWo3|L0J4~Cm}^)i8B!P}8AKQ?7?>GsK%y*g`N=~37L2}=z4=2~nH3n7
zm?tmgcU7rjDdMSNsbS1!DM~A0Sio4qkj1osIfb!?VIk8(#uO$=hFaFid;*S)43h%{
a4)9L_xkd;`FmUoPN-%Af7vx}M<N*Nep(Ymq

delta 241
zcmeBCS)sxg%FD~e00gnG`{KLBHuBBltLFf6oq@Qx3`mqP)G#zN*0Pi^)-W|Qx-i5h
z)UwvFWHHsSvNA|AlrYyY)-czwN;0G{N-_XdFe7*(3>FN`3^oiv)hq;5Gf6VQR2K?N
z=H^#s^qZ{DAIi$8$7sMbxs%^jrG#MtV+lhR(*ouc#u|o&ObZ!PfSN!~DN?CnsbS1!
lDN39CiQkb?VzP<A0se^~=L-P|1`Y`(9>&egf*g#DoB*StF6#gQ

diff --git a/rgbAI/lib/ailib/ai.py b/rgbAI/lib/ailib/ai.py
index afa9348..cd3d0cb 100644
--- a/rgbAI/lib/ailib/ai.py
+++ b/rgbAI/lib/ailib/ai.py
@@ -137,8 +137,11 @@ def mutateProps( inpObj, curCost:float, maxLen:int, gradient:list ):
 	obj = copy(inpObj)
 
 	for i in range(maxLen):
-		obj.weights[i] -= getLearningRate( curCost, gradient[i]["weight"], maxLen ) * gradient[i]["weight"] # mutate the weights
-		obj.bias[i] -= getLearningRate( curCost, gradient[i]["weight"], maxLen ) * gradient[i]["bias"]
+		# obj.weights[i] -= getLearningRate( curCost, gradient[i]["weight"], maxLen ) * gradient[i]["weight"] # mutate the weights
+		# obj.bias[i] -= getLearningRate( curCost, gradient[i]["weight"], maxLen ) * gradient[i]["bias"]
+		obj.weights[i] -= obj.learningrate * gradient[i]["weight"] # mutate the weights
+		obj.bias[i] -= obj.learningrate * gradient[i]["bias"]
+
 
 	return obj
 
@@ -150,9 +153,10 @@ def learn( inputNum:int, targetCost:float, obj, theta:float, curCost: float=None
 	# i.e. : W' = W - lr * gradient (respect to W in layer i) = W - lr*[ dC / dW[i] ... ]
 	# So if we change all the weights with i.e. 0.01 = theta, then we can derive the gradient with math and stuff
 
-	inp = np.asarray(np.random.rand( 1, inputNum ))[0] # create a random learning sample
-
+	
 	while( not curCost or curCost > targetCost ): # targetCost is the target for the cost function
+		inp = np.asarray(np.random.rand( 1, inputNum ))[0] # create a random learning sample
+
 		maxLen = len(obj.bias)
 		grads, costW, costB, curCost = gradient( inp, obj, theta, maxLen - 1 )