for j in range(8):
z[hid[j]] = v[hid[j]]
c = np.matmul(z,w)
z = np.zeros(60)
display.display(c,28,28)
m = np.mean(data[:500],axis=0)
c = np.matmul((b+v),w)
'''
#display.display(c,28,28)
#display.display(np.ndarray.flatten(np.array(m)),28,28)
if(int(options.save)):
for i in range(50):
display.save(data[i],28,28,folder='./imgs',name='in',index=i)
res = auto.get_output(np.asarray([data[i]]),session=session)
#if(options.normed):
# res = res + np.cumsum(data,axis=0)[-1]/data.shape[0]
display.save(res,28,28,folder='./imgs',name='out',index=i)
if(int(options.classifier)):
h = auto.get_hidden(np.asarray(data),session=session)
k = knn_classifier.knn_classifier(data=h,label=lab,k=3)
k.learn()
test, labtest = get_data_from_minst.get_test_from_mnist()
t = auto.get_hidden(np.asarray(test),session=session)
hid = [0,1,27,28,39,45,52,53]
for j in range(8):
z[hid[j]] = v[hid[j]]
c = np.matmul(z,w)
z = np.zeros(60)
display.display(c,28,28)
m = np.mean(data[:500],axis=0)
c = np.matmul((b+v),w)'''
#display.display(c,28,28)
#display.display(np.ndarray.flatten(np.array(m)),28,28)
if(int(options.save)):
for i in range(50):
display.save(data[i],70,80,folder='./imgs_pie',name='in',index=i)
res = auto.get_output(np.asarray([data[i]]),session=session)
display.save(res,70,80,folder='./imgs_pie',name='out',index=i)
if(int(options.classifier)):
train = np.loadtxt("../datasets/multi_pie_train.dat")
l_train = np.loadtxt("../datasets/multi_pie_l_train.dat")
test = np.loadtxt("../datasets/multi_pie_test.dat")
l_test = np.loadtxt("../datasets/multi_pie_l_test.dat")
train = train+abs(np.min(train))
train = train/np.max(train)
train = train.astype("float32")
act = ['tanh','tanh','tanh']
act2 = ['linear','linear','linear']
auto = autoencoder(units,act)
auto.generate_encoder(euris=True)
auto.generate_decoder(symmetric=False,act=act2)
session = auto.init_network()
#bat = batch.rand_batch(data,70)
ic,bc = auto.train(data,batch=None,n_iters=5000,use_dropout=False,keep_prob=0.5,reg_weight=False,reg_lambda=0.0,model_name='conv_7.mod',display=False,noise=False,gradient='adam',learning_rate=0.000125)
print "Finished. Initial cost: ",ic," Final cost: ",bc
auto.load_model('conv_7.mod',session=session)
#data = data+m
auto.stop_dropout()
red_feat = auto.get_hidden(data,session=session)
out_feat = auto.get_output(data,session=session)
np.savetxt("red_7",red_feat)
np.savetxt("out_7",out_feat)
nd = np.transpose(data)
no = np.transpose(out_feat)
if(sys.argv[1]=='s'):
for i in range(64):
#display.save(nd[i],3,3,3,folder='./feat',name='in_12',index=i)
display.save(no[i],3,3,3,folder='./feat',name='out_7',index=i)