learning_rate = 0.0001
batch_size = 1000
n_modalities = 5
size_modalities = [4, 4, 4, 4, 1, 1, 1, 1, 4, 4]
numModels = [10, 10, 10, 10, 2, 2, 2, 2, 10, 10]
numFactors = [10, 10, 10, 10, 2, 2, 2, 2, 10, 10]
used_modalities = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
corruption_level = 0.0
softmaxnoise = 0.0
numClass = 1
numParam = 100
vanilla = True
network = arch(n_modalities, size_modalities, numModels,
numFactors, numClass, numParam, used_modalities,
batch_size, learning_rate, vanilla,
corruption_level, softmaxnoise)
new_saver = tf.train.Saver()
new_saver.restore(sess, "./models/droniou_complete.ckpt")
print("Model restored.")
if (test_id == 1):
test_1()
if (test_id == 2):
test_2()
if (test_id == 3):
test_3()
if (test_id == 4):
test_4()
with tf.Graph().as_default() as g:
with tf.Session() as sess:
# Network parameters
learning_rate = 0.0001
batch_size = 1000
n_modalities=5
size_modalities=[8,8,2,2,8]
numModels=[20,20,5,5,20]
numFactors=[20,20,5,5,20]
used_modalities = np.zeros(n_modalities)
#used_modalities =
numClass=1
numParam=100
network = arch( n_modalities, size_modalities, numModels, numFactors, numClass, numParam, used_modalities, batch_size,learning_rate)
with tf.Session() as sess:
new_saver = tf.train.Saver()
new_saver.restore(sess, "./models_vanilla/droniou_complete.ckpt")
print("Model restored.")
# Test 1: complete data
print('Test 1')
sample_init = 100
x_sample = X_augm_test[sample_init:sample_init+batch_size,:28]
x_reconstruct = network.reconstruct(sess, x_sample)
scipy.io.savemat("results/mvae_final_test1.mat",{"x_reconstruct":x_reconstruct,"x_sample":x_sample})