def main(unused_argv):
FLAGS.logdir = os.path.join(FLAGS.logdir, FLAGS.algo)
g = tf.Graph()
with g.as_default():
data_batch, _, _ = mnist_data.get_mnist(batch_size=FLAGS.batch_size,
split="train")
if FLAGS.algo == "nis_vae_proposal":
print("Running NIS with VAE proposal")
elbo = make_nis_with_vae_proposal(data_batch,
K=FLAGS.K,
vae_latent_dim=FLAGS.latent_dim,
nis_hidden_sizes=[200, 100],
q_hidden_sizes=[300, 300],
p_x_hidden_sizes=[300, 300],
scale_min=FLAGS.scale_min)
elif FLAGS.algo == "nis_gaussian_proposal":
print("Running NIS with Gaussian proposal")
elbo = make_nis_with_gaussian_proposal(
data_batch, K=FLAGS.K, hidden_layer_sizes=[200, 100])
elif FLAGS.algo == "vae_nis_prior":
print("Running VAE with NIS prior")
elbo = make_vae_with_nis_prior(data_batch,
latent_dim=FLAGS.latent_dim,
K=FLAGS.K,
nis_hidden_sizes=[200, 100],
q_hidden_sizes=[300, 300],
p_x_hidden_sizes=[300, 300],
scale_min=FLAGS.scale_min)
elif FLAGS.algo == "vae_gaussian_prior":
print("Running VAE with gaussian prior")
elbo = make_vae_with_gaussian_prior(data_batch,
latent_dim=FLAGS.latent_dim,
q_hidden_sizes=[300, 300],
p_x_hidden_sizes=[300, 300],
scale_min=1e-5)
# Finish constructing the graph
elbo_avg = tf.reduce_mean(elbo)
tf.summary.scalar("elbo", elbo_avg)
global_step = tf.train.get_or_create_global_step()
opt = tf.train.AdamOptimizer(learning_rate=FLAGS.learning_rate)
grads = opt.compute_gradients(-elbo_avg)
train_op = opt.apply_gradients(grads, global_step=global_step)
log_hooks = make_log_hooks(global_step, elbo_avg)
with tf.train.MonitoredTrainingSession(
master="",
is_chief=True,
hooks=log_hooks,
checkpoint_dir=FLAGS.logdir,
save_checkpoint_secs=120,
save_summaries_steps=FLAGS.summarize_every,
log_step_count_steps=FLAGS.summarize_every) as sess:
cur_step = -1
while cur_step <= FLAGS.max_steps and not sess.should_stop():
_, cur_step = sess.run([train_op, global_step])
import numpy as np
import logging
import mnist_data as data
from math import sqrt
from autoencoder import AutoEncoderModel
if __name__ == '__main__':
lv = 1e-2# lv/ln in CDL
# set to INFO to see less information during training
logging.basicConfig(level=logging.DEBUG)
#ae_model = AutoEncoderModel(mx.gpu(0), [784,500,500,2000,10], pt_dropout=0.2,
# internal_act='relu', output_act='relu')
ae_model = AutoEncoderModel(mx.cpu(2), [784,500,500,2000,10], pt_dropout=0.2,
internal_act='relu', output_act='relu')
X, _ = data.get_mnist()
train_X = X[:60000]
val_X = X[60000:]
#ae_model.layerwise_pretrain(train_X, 256, 50000, 'sgd', l_rate=0.1, decay=0.0,
# lr_scheduler=mx.misc.FactorScheduler(20000,0.1))
#V = np.zeros((train_X.shape[0],10))
V = np.random.rand(train_X.shape[0],10)/10
lambda_v_rt = np.ones((train_X.shape[0],10))*sqrt(lv)
ae_model.finetune(train_X, V, lambda_v_rt, 256,
20, 'sgd', l_rate=0.1, decay=0.0,
lr_scheduler=mx.misc.FactorScheduler(20000,0.1))
ae_model.save('mnist_pt.arg')
ae_model.load('mnist_pt.arg')
print "Training error:", ae_model.eval(train_X,V,lambda_v_rt)
#print "Validation error:", ae_model.eval(val_X)
from snn_multipleneurons_fast import *
import sys
import os
from sklearn.svm import LinearSVC
from mnist_data import get_mnist
import argparse
from tensorboardX import SummaryWriter
from tqdm import tqdm
import warnings
warnings.filterwarnings("ignore")
run = input("Run folder: ")
x_train, x_test, y_train, y_test = get_mnist()
clf_output = LinearSVC(tol=1e-6, dual=False, class_weight='balanced')
if run == 'raw-mnist':
print("Doing MNIST")
tic = time.time()
clf_output.fit(x_train, y_train)
train_score = clf_output.score(x_train, y_train)
test_score = clf_output.score(x_test, y_test)
print("Time elapsed", time.time() - tic)
print("MNIST baseline:")
print('Test Score: ', test_score)
print('Train Score: ', train_score, flush=True)
else: