def rps_classify():
truth_in = {}
truth_out = {}
net.loadWeights("rps_weigths.txt")
truth_in[0] = i_o.get_converted_img('nao', 0)
net.calcOuts(truth_in[0])
print net.outs[2]
for i in range(len(net.outs[2])):
if net.outs[2][i] > 0.5:
idx = net.outs[2].index(max(net.outs[2]))
return directories[idx]
#return failure otherwise
return 'nothing'
def rps_learn():
truth_in = {}
truth_out = {}
z = 0
chosen = [0] * 4
for z in range(0,75):
#random dir
dir = directories[randint(0,3)]
#import all pictures in dir.
for root, dirs, files in os.walk('pics/'+dir):
print(len(files))
file = files[randint(0,len(files)-1)]
bestand = 'pics/'+dir+'/'+file
truth_in[z] = i_o.get_converted_img(bestand, 1)
#print bestand
correct_out = {}
#set correct outputs
if dir == 'rock':
correct_out = [1,0,0,0]
chosen[0] += 1
if dir == 'paper':
correct_out = [0,1,0,0]
chosen[1] += 1
if dir == 'scissors':
correct_out = [0,0,1,0]
chosen[2] += 1
if dir == 'green':
correct_out = [0,0,0,1]
chosen[3] += 1
truth_out[z] = correct_out
#print truth_in[z]
#print truth_out[z]
print('done', len(files))
print 'chosen images:'
print chosen
char = raw_input('continue learning with this test set? Y or N')
if char == 'n':
print('exiting program')
return
#truth_in = [[0,0],[0,1],[1,0],[1,1]]
#truth_out = [[0],[1],[1],[0]]
### Neural Network ###
#net = network(18,1,10,3) #inputs, hidden_layers, hidden_neurons, outputs
net.initWeights()
#net.loadWeights("wgts_18,11,4_448_2,0.txt")
net.debug = False
net.alpha = 1 #Learning rate
net.adaptive_alpha = True
net.alpha_roof = 1
#net.calcOuts(truth_in[0])
#print net.outs[2]
net.useGraph()
net.graph = True
net.graphFreq = 1
#net.train(truth_in,truth_out,0,50) #input_set, output_set, learning_rate, mode, epochs
#print "Training"
cnt = net.train(truth_in,truth_out,1,3) #input_set, output_set, learning_rate, mode, target_sse
net.saveWeights("rps_weigths.txt")
#print "Saved weights"
print('done')
#net.showNet(True,0)
#a = i_o.get_converted_img('pics/globe.jpg')
#print a
#net.calcOuts(a)
#print net.outs[1]
net.calcOuts(truth_in[0])
print net.outs[2]
