def autoex1(epochs=2000, num_bits=3, lrate=0.5, tint=25, showint=100):
ann = autoencoder(nh=num_bits, lr=lrate)
PLT.ion()
ann.do_training(epochs, test_interval=tint, show_interval=showint)
ann.do_testing(
scatter=True
) # Do a final round of testing to plot the hidden-layer activation vectors.
PLT.ioff()
TFT.close_session(ann.current_session)
return ann
def tfex7(n=5):
w = tf.Variable(np.random.uniform(-1, 1, size=(5, 5)),
name='w') # Same matrix as before
x = tf.Variable(np.zeros((1, 5)), name='x')
y = tf.placeholder(tf.float64, shape=(1, 5), name='y')
feeder = {y: np.random.uniform(-1, 1, size=(1, 5))}
update_x = x.assign(tf.matmul(x, w) + y)
_, _, sess = quickrun4([update_x], [w, x, y], feed_dict=feeder)
for step in range(n):
quickrun4([update_x], [x], session=sess, feed_dict=feeder)
TFT.close_session(sess)
def tfex8(size=5, steps=50, tvect=None, learning_rate=0.5, showint=10):
target = tvect if tvect else np.ones((1, size))
w = tf.Variable(np.random.uniform(-.1, .1, size=(size, size)),
name='weights') # weights applied to x.
b = tf.Variable(np.zeros((1, size)), name='bias') # bias terms
x = tf.placeholder(tf.float64, shape=(1, size), name='input')
y = tf.sigmoid(
tf.matmul(x, w) + b, name='out-sigmoid'
) # Gather all weighted inputs, then apply activation function
error = tf.reduce_mean(tf.square(target - y))
optimizer = tf.train.GradientDescentOptimizer(learning_rate)
training_operator = optimizer.minimize(error)
feeder = {x: np.random.uniform(-1, 1, size=(1, size))}
sess = TFT.gen_initialized_session()
for step in range(steps):
quickrun4([training_operator], [w, b, y],
session=sess,
feed_dict=feeder,
step=step,
show_interval=showint)
TFT.close_session(sess)
def autoex1(epochs=1000000,
num_bits=15,
lrate=1,
tint=100,
showint=10000,
mbs=53):
#case_generator = (lambda: TFT.gen_all_parity_cases(10))
#case_generator = (lambda: TFT.gen_segmented_vector_cases(25,1000,0,8))
#case_generator = (lambda: TFT.gen_vector_count_cases(500,15))
case_generator = "yeast.txt"
cman = Caseman(cfunc=case_generator,
sep=",",
vfrac=0,
tfrac=0,
stdeviation=True)
ann = autoencoder(nh=num_bits, lr=lrate, cman=cman, mbs=mbs)
PLT.ion()
ann.do_training(epochs, test_interval=tint, show_interval=showint, mbs=mbs)
ann.do_testing(
scatter=True, mbs=mbs
) # Do a final round of testing to plot the hidden-layer activation vectors.
PLT.ioff()
TFT.close_session(ann.current_session, False)
return ann
def close_current_session(self, view=True):
self.save_session_params(sess=self.current_session)
TFT.close_session(self.current_session, view=view)
def mainfunc(steps=1000000,
lrate="scale",
tint=100,
showint=10000,
mbs=100,
wgt_range=(-.3, .3),
hidden_layers=[50, 50],
hidac=(lambda x, y: tf.tanh(x, name=y)),
outac=(lambda x, y: tf.nn.softmax(x, name=y)),
case_generator="mnist.txt",
stdeviation=False,
vfrac=0.1,
tfrac=0.1,
cfunc="rmse",
mapbs=0,
map_layers=[],
display_wgts=[],
display_biases=[],
cfrac=1.0,
dendrogram_layers=[],
numeric=False):
cman = Caseman(cfunc=case_generator,
cfrac=cfrac,
vfrac=vfrac,
tfrac=tfrac,
stdeviation=stdeviation)
ann = Gann(lr=lrate,
cman=cman,
mbs=mbs,
wgt_range=wgt_range,
hidden_layers=hidden_layers,
hidac=hidac,
outac=outac,
cfunc=cfunc,
dendrogram_layers=dendrogram_layers)
PLT.ion()
epochs = int(steps / mbs)
ann.do_training(epochs, test_interval=tint, show_interval=showint, mbs=mbs)
for out_layer in map_layers:
ann.add_mapvar(out_layer - 1, "out")
for wgts_layer in display_wgts:
if (wgts_layer == 0):
continue
ann.add_mapvar(wgts_layer - 1, "wgt")
for bias_layer in display_biases:
if (bias_layer == 0):
continue
ann.add_mapvar(bias_layer - 1, "bias")
for dendro_layer in dendrogram_layers:
if (dendro_layer == 0):
continue
ann.add_mapvar(dendro_layer - 1, type="out")
##OBS: Skal vel ikke vise dendrogram mellom input og output i hele nettverket, men for et ønsket layer
if (mapbs > 0):
ann.do_mapping(mbs=mbs, mapbs=mapbs, numeric=numeric)
PLT.ioff()
TFT.close_session(ann.current_session, False)
return ann
def closeCurrentSession(self, view = True): self.saveSessionParams(sess = self.currentSession) tft.close_session(self.currentSession, view = view)
