def simulacao(input_dim,
num_samples,
sparseness,
learning_rate,
model_seed,
data_seed,
nsteps=1000,
max_count=100):
acc = np.zeros(nsteps + 1)
myDataSet = buildRandomBoolDataSet(num_samples, input_dim,
seed=data_seed).batch(num_samples)
myIter = myDataSet.make_initializable_iterator()
myFeatures, myLabels = myIter.get_next()
x = myFeatures
y_ = myLabels
model = Model(x, y_, sparseness, learning_rate, seed=model_seed)
model.set_vanilla_loss()
# initialize variables
sess.run(tf.global_variables_initializer())
sess.run(myIter.initializer)
acc[0] = model.accuracy.eval()
count = 1
for ss in np.arange(1, nsteps + 1):
sess.run(myIter.initializer)
model.reduce_sparsness()
model.train_step.run()
model.increase_sparsness()
sess.run(myIter.initializer)
acc[ss] = model.accuracy.eval()
if (acc[ss] == acc[ss - count]):
count += 1
else:
count = 1
if (count == max_count):
acc[ss:] = acc[ss]
break
return acc
y_ = myLabels
#model = Model(x, y_, spars, 0.1) # simple 2-layer network
#print("data_seed")
#print(data_seed)
model = Model(x, y_, sparseness, learning_rate,
seed=data_seed) # simple 2-layer network
model.set_vanilla_loss()
# initialize variables
sess.run(tf.global_variables_initializer())
sess.run(myIter.initializer)
#acc[0] = model.accuracy.eval()
count = 1
for ss in np.arange(1, nsteps + 1):
sess.run(myIter.initializer)
model.reduce_sparsness()
model.train_step.run()
#model.aumenta spar
sess.run(myIter.initializer)
#acc[ss] = model.accuracy.eval()
acc[zz, ii, jj] = model.accuracy.eval()
sess.close()
# %%
ax = plt.subplot(111)
#plt.plot(acc[:,:,0])
for zz, sparseness in enumerate(sparse):
plt.plot(num_samplesV,