def main():
parser = make_standard_parser('Random Projection RL experiments.',
arch_choices=arch_choices,
skip_train=True,
skip_val=True)
parser.add_argument('--vsize',
type=int,
default=100,
help='Dimension of intrinsic parmaeter space.')
parser.add_argument('--depth',
type=int,
default=2,
help='Number of layers in FNN.')
parser.add_argument('--width',
type=int,
default=100,
help='Width of layers in FNN.')
parser.add_argument('--env_name',
type=str,
default=arch_choices[0],
choices=env_choices,
help='Which architecture to use (choices: %s).' %
env_choices)
args = parser.parse_args()
proj_type = 'dense'
if args.arch == 'fourier':
d_Fourier = 200
train_style, val_style = ('',
'') if args.nocolor else (colorama.Fore.BLUE,
colorama.Fore.MAGENTA)
# Get a TF session registered with Keras and set numpy and TF seeds
sess = setup_session_and_seeds(args.seed)
# 0. LOAD ENV
theta_r = 2.0
theta_threshold_radians = 12 * 2 * math.pi / 360
x_threshold = 1
env_name = args.env_name
env = gym.make(env_name)
#sess = tf.Session()
state_dim = env.observation_space.shape[0]
if env_name == 'Pendulum-v0':
num_actions = 1
else:
num_actions = env.action_space.n
# 1. CREATE MODEL
extra_feed_dict = {}
with WithTimer('Make model'):
if args.arch == 'fc_dir':
model = build_model_fc_dir(state_dim,
num_actions,
weight_decay=args.l2,
depth=args.depth,
width=args.width,
shift_in=None)
elif args.arch == 'fc':
model = build_model_fc(state_dim,
num_actions,
weight_decay=args.l2,
vsize=args.vsize,
depth=args.depth,
width=args.width,
shift_in=None,
proj_type='dense')
elif args.arch == 'fourier':
model = build_model_fourier(d_Fourier,
num_actions,
weight_decay=args.l2,
depth=args.depth,
width=args.width,
shift_in=None)
else:
raise Exception('Unknown network architecture: %s' % args.arch)
print 'All model weights:'
summarize_weights(model.trainable_weights)
print 'Model summary:'
model.summary()
model.print_trainable_warnings()
# 2. COMPUTE GRADS AND CREATE OPTIMIZER
if args.opt == 'sgd':
optimizer = tf.train.MomentumOptimizer(args.lr, args.mom)
elif args.opt == 'rmsprop':
optimizer = tf.train.RMSPropOptimizer(args.lr, momentum=args.mom)
elif args.opt == 'adam':
optimizer = tf.train.AdamOptimizer(args.lr, args.beta1, args.beta2)
summarize_opt(optimizer)
# 3. OPTIONALLY SAVE OR LOAD VARIABLES (e.g. model params, model running BN means, optimization momentum, ...) and then finalize initialization
saver = tf.train.Saver(
max_to_keep=None) if (args.output or args.load) else None
if args.load:
ckptfile, miscfile = args.load.split(':')
# Restore values directly to graph
saver.restore(sess, ckptfile)
with gzip.open(miscfile) as ff:
saved = pickle.load(ff)
buddy = saved['buddy']
else:
buddy = StatsBuddy()
buddy.tic() # call if new run OR resumed run
# Initialize any missed vars (e.g. optimization momentum, ... if not loaded from checkpoint)
uninitialized_vars = tf_get_uninitialized_variables(sess)
init_missed_vars = tf.variables_initializer(uninitialized_vars,
'init_missed_vars')
sess.run(init_missed_vars)
# Print warnings about any TF vs. Keras shape mismatches
warn_misaligned_shapes(model)
# Make sure all variables, which are model variables, have been initialized (e.g. model params and model running BN means)
tf_assert_all_init(sess)
# 3.5 Normalize the overall basis matrix across the (multiple) unnormalized basis matrices for each layer
basis_matrices = []
normalizers = []
for layer in model.layers:
try:
basis_matrices.extend(layer.offset_creator.basis_matrices)
except AttributeError:
continue
try:
normalizers.extend(layer.offset_creator.basis_matrix_normalizers)
except AttributeError:
continue
if len(basis_matrices) > 0 and not args.load:
if proj_type == 'sparse':
# Norm of overall basis matrix rows (num elements in each sum == total parameters in model)
bm_row_norms = tf.sqrt(
tf.add_n([
tf.sparse_reduce_sum(tf.square(bm), 1)
for bm in basis_matrices
]))
# Assign `normalizer` Variable to these row norms to achieve normalization of the basis matrix
# in the TF computational graph
rescale_basis_matrices = [
tf.assign(var, tf.reshape(bm_row_norms, var.shape))
for var in normalizers
]
_ = sess.run(rescale_basis_matrices)
elif proj_type == 'dense':
bm_sums = [
tf.reduce_sum(tf.square(bm), 1) for bm in basis_matrices
]
divisor = tf.expand_dims(tf.sqrt(tf.add_n(bm_sums)), 1)
rescale_basis_matrices = [
tf.assign(var, var / divisor) for var in basis_matrices
]
_ = sess.run(rescale_basis_matrices)
else:
print '\nhere\n'
embed()
assert False, 'what to do with fastfood?'
# 3.5 Fourier features of the observations if required
if args.arch == 'fourier':
RP = np.random.randn(d_Fourier, state_dim)
Rb = np.random.uniform(-math.pi, math.pi, d_Fourier)
state_dim = d_Fourier
# 4. SETUP TENSORBOARD LOGGING
writer = None
if args.output:
mkdir_p(args.output)
writer = tf.summary.FileWriter(
args.output + "/{}-experiment-1".format(env_name), sess.graph)
observation_to_action = model
# 5. TRAIN
if env_name == 'CartPole-v0':
q_learner = NeuralQLearner(
sess,
optimizer,
observation_to_action,
state_dim,
num_actions,
init_exp=0.5,
anneal_steps=10000, # N steps for annealing exploration
discount_factor=0.95,
batch_size=32,
target_update_rate=0.01,
summary_writer=writer)
MAX_EPISODES = 10000
MAX_STEPS = 200
episode_history = deque(maxlen=100)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
episode_history = deque(maxlen=100)
for i_episode in range(MAX_EPISODES):
# initialize
state = env.reset()
if args.arch == 'fourier':
state = np.sin(RP.dot(state) +
Rb) # np.median( RP.dot(state) )
total_rewards = 0
for t in range(MAX_STEPS):
# env.render()
action = q_learner.eGreedyAction(state[np.newaxis, :])
next_state, reward, done, _ = env.step(action)
if args.arch == 'fourier':
next_state = np.sin(RP.dot(next_state) +
Rb) # np.median( RP.dot(state) )
total_rewards += reward
# reward = -10 if done else 0.1 # normalize reward
q_learner.storeExperience(state, action, reward, next_state,
done)
q_learner.updateModel()
state = next_state
if done: break
episode_history.append(total_rewards)
mean_rewards = np.mean(episode_history)
print("Episode {}".format(i_episode))
print("Finished after {} timesteps".format(t + 1))
print("Reward for this episode: {}".format(total_rewards))
print("Average reward for last 100 episodes: {:.2f}".format(
mean_rewards))
if mean_rewards >= 195.0:
print("Environment {} solved after {} episodes".format(
env_name, i_episode + 1))
break
elif env_name == 'CartPole-v1':
q_learner = NeuralQLearner(
sess,
optimizer,
observation_to_action,
state_dim,
num_actions,
init_exp=0.5,
anneal_steps=10000, # N steps for annealing exploration
discount_factor=0.95,
batch_size=32,
target_update_rate=0.01,
summary_writer=writer)
MAX_EPISODES = 10000
MAX_STEPS = 500
episode_history = deque(maxlen=100)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
episode_history = deque(maxlen=100)
for i_episode in range(MAX_EPISODES):
# initialize
state = env.reset()
total_rewards = 0
for t in itertools.count():
# env.render()
action = q_learner.eGreedyAction(state[np.newaxis, :])
next_state, reward, done, _ = env.step(action)
total_rewards += reward
# reward = -10 if done else 0.1 # normalize reward
q_learner.storeExperience(state, action, reward, next_state,
done)
q_learner.updateModel()
state = next_state
if done: break
episode_history.append(total_rewards)
mean_rewards = np.mean(episode_history)
print("Episode {}".format(i_episode))
print("Finished after {} timesteps".format(t + 1))
print("Reward for this episode: {}".format(total_rewards))
print("Average reward for last 100 episodes: {:.2f}".format(
mean_rewards))
if mean_rewards >= 475.0:
print("Environment {} solved after {} episodes".format(
env_name, i_episode + 1))
break
elif env_name == 'MountainCar-v0':
q_learner = NeuralQLearner(
sess,
optimizer,
observation_to_action,
state_dim,
num_actions,
batch_size=64,
anneal_steps=10000, # N steps for annealing exploration
replay_buffer_size=1000000,
discount_factor=0.95, # discount future rewards
target_update_rate=0.01,
reg_param=0.01, # regularization constants
summary_writer=writer)
MAX_EPISODES = 10000
episode_history = deque(maxlen=100)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
episode_history = deque(maxlen=100)
for i_episode in range(MAX_EPISODES):
# initialize
state = env.reset()
total_rewards = 0
for t in itertools.count():
# env.render()
action = q_learner.eGreedyAction(state[np.newaxis, :])
next_state, reward, done, _ = env.step(action)
total_rewards += reward
# reward = -10 if done else 0.1 # normalize reward
q_learner.storeExperience(state, action, reward, next_state,
done)
q_learner.updateModel()
state = next_state
if done: break
episode_history.append(total_rewards)
mean_rewards = np.mean(episode_history)
print("Episode {}".format(i_episode))
print("Finished after {} timesteps".format(t + 1))
print("Reward for this episode: {}".format(total_rewards))
print("Average reward for last 100 episodes: {:.2f}".format(
mean_rewards))
if mean_rewards >= -110.0:
print("Environment {} solved after {} episodes".format(
env_name, i_episode + 1))
break
elif env_name == 'CartPole-v2' or env_name == 'CartPole-v3' or env_name == 'CartPole-v4' or env_name == 'CartPole-v5':
q_learner = NeuralQLearner(
sess,
optimizer,
observation_to_action,
state_dim,
num_actions,
init_exp=0.5,
anneal_steps=10000, # N steps for annealing exploration
discount_factor=0.95,
batch_size=32,
target_update_rate=0.01,
summary_writer=writer)
MAX_EPISODES = 10000
MAX_STEPS = 200
episode_history = deque(maxlen=100)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
episode_history = deque(maxlen=100)
for i_episode in range(MAX_EPISODES):
# initialize
state = env.reset()
if args.arch == 'fourier':
state = np.sin(RP.dot(state) +
Rb) # np.median( RP.dot(state) )
total_rewards = 0
for t in range(MAX_STEPS):
# env.render()
action = q_learner.eGreedyAction(state[np.newaxis, :])
next_state, reward, done, _ = env.step(action)
if args.arch == 'fourier':
next_state = np.sin(RP.dot(next_state) +
Rb) # np.median( RP.dot(state) )
total_rewards += reward
# reward = -10 if done else 0.1 # normalize reward
q_learner.storeExperience(state, action, reward, next_state,
done)
q_learner.updateModel()
state = next_state
if done: break
episode_history.append(total_rewards)
mean_rewards = np.mean(episode_history)
print("Episode {}".format(i_episode))
print("Finished after {} timesteps".format(t + 1))
print("Reward for this episode: {}".format(total_rewards))
print("Average reward for last 100 episodes: {:.2f}".format(
mean_rewards))
if mean_rewards >= 195.0:
print("Environment {} solved after {} episodes".format(
env_name, i_episode + 1))
break
elif env_name == 'Pendulum-v0':
q_learner = NeuralQLearner(
sess,
optimizer,
observation_to_action,
state_dim,
num_actions,
init_exp=0.5,
anneal_steps=10000, # N steps for annealing exploration
discount_factor=0.99,
batch_size=32,
target_update_rate=0.01,
summary_writer=writer)
MAX_EPISODES = 10000
MAX_STEPS = 200
episode_history = deque(maxlen=100)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
episode_history = deque(maxlen=100)
for i_episode in range(MAX_EPISODES):
# initialize
state = env.reset()
if args.arch == 'fourier':
state = np.sin(RP.dot(state) +
Rb) # np.median( RP.dot(state) )
total_rewards = 0
for t in range(MAX_STEPS):
# env.render()
action = q_learner.eGreedyAction(state[np.newaxis, :])
next_state, reward, done, _ = env.step(action)
if args.arch == 'fourier':
next_state = np.sin(RP.dot(next_state) +
Rb) # np.median( RP.dot(state) )
total_rewards += reward
# reward = -10 if done else 0.1 # normalize reward
q_learner.storeExperience(state, action, reward, next_state,
done)
q_learner.updateModel()
state = next_state
if done: break
episode_history.append(total_rewards)
mean_rewards = np.mean(episode_history)
print("Episode {}".format(i_episode))
print("Finished after {} timesteps".format(t + 1))
print("Reward for this episode: {}".format(total_rewards))
print("Average reward for last 100 episodes: {:.2f}".format(
mean_rewards))
if mean_rewards >= 195.0:
print("Environment {} solved after {} episodes".format(
env_name, i_episode + 1))
break
def main():
parser = make_standard_parser(
'Coordconv',
arch_choices=arch_choices,
skip_train=True,
skip_val=True)
# re-add train and val h5s as optional
parser.add_argument('--data_h5', type=str,
default='./data/rectangle_4_uniform.h5',
help='data file in hdf5.')
parser.add_argument('--x_dim', type=int, default=64,
help='x dimension of the output image')
parser.add_argument('--y_dim', type=int, default=64,
help='y dimension of the output image')
parser.add_argument('--lrpolicy', type=str, default='constant',
choices=lr_policy_choices, help='LR policy.')
parser.add_argument('--lrstepratio', type=float,
default=.1, help='LR policy step ratio.')
parser.add_argument('--lrmaxsteps', type=int, default=5,
help='LR policy step ratio.')
parser.add_argument('--lrstepevery', type=int, default=50,
help='LR policy step ratio.')
parser.add_argument('--filter_size', '-fs', type=int, default=3,
help='filter size in deconv network')
parser.add_argument('--channel_mul', '-mul', type=int, default=2,
help='Deconv model channel multiplier to make bigger models')
parser.add_argument('--use_mse_loss', '-mse', action='store_true',
help='use mse loss instead of cross entropy')
parser.add_argument('--use_sigm_loss', '-sig', action='store_true',
help='use sigmoid loss instead of cross entropy')
parser.add_argument('--interm_loss', '-interm', default=None,
choices=(None, 'softmax', 'mse'),
help='add intermediate loss to end-to-end painter model')
parser.add_argument('--no_softmax', '-nosfmx', action='store_true',
help='Remove softmax sharpening layer in model')
args = parser.parse_args()
if args.lrpolicy == 'step':
lr_policy = LRPolicyStep(args)
elif args.lrpolicy == 'valstep':
lr_policy = LRPolicyValStep(args)
else:
lr_policy = LRPolicyConstant(args)
minibatch_size = args.minibatch
train_style, val_style = (
'', '') if args.nocolor else (
colorama.Fore.BLUE, colorama.Fore.MAGENTA)
sess = setup_session_and_seeds(args.seed, assert_gpu=not args.cpu)
# 0. Load data or generate data on the fly
print 'Loading data: {}'.format(args.data_h5)
if args.arch in ['deconv_classification',
'coordconv_classification',
'upsample_conv_coords',
'upsample_coordconv_coords']:
# option a: generate data on the fly
#data = list(itertools.product(range(args.x_dim),range(args.y_dim)))
# random.shuffle(data)
#train_test_split = .8
#val_reps = int(args.x_dim * args.x_dim * train_test_split) // minibatch_size
#val_size = val_reps * minibatch_size
#train_end = args.x_dim * args.x_dim - val_size
#train_x, val_x = np.array(data[:train_end]).astype('int'), np.array(data[train_end:]).astype('int')
#train_y, val_y = None, None
#DATA_GEN_ON_THE_FLY = True
# option b: load the data
fd = h5py.File(args.data_h5, 'r')
train_x = np.array(fd['train_locations'], dtype=int) # shape (2368, 2)
train_y = np.array(fd['train_onehots'], dtype=float) # shape (2368, 64, 64, 1)
val_x = np.array(fd['val_locations'], dtype=float) # shape (768, 2)
val_y = np.array(fd['val_onehots'], dtype=float) # shape (768, 64, 64, 1)
DATA_GEN_ON_THE_FLY = False
# number of image channels
image_c = train_y.shape[-1] if train_y is not None and len(train_y.shape) == 4 else 1
elif args.arch == 'conv_onehot_image':
fd = h5py.File(args.data_h5, 'r')
train_x = np.array(
fd['train_onehots'],
dtype=int) # shape (2368, 64, 64, 1)
train_y = np.array(fd['train_imagegray'],
dtype=float) / 255.0 # shape (2368, 64, 64, 1)
val_x = np.array(
fd['val_onehots'],
dtype=float) # shape (768, 64, 64, 1)
val_y = np.array(fd['val_imagegray'], dtype=float) / \
255.0 # shape (768, 64, 64, 1)
image_c = train_y.shape[-1]
elif args.arch == 'deconv_rendering':
fd = h5py.File(args.data_h5, 'r')
train_x = np.array(fd['train_locations'], dtype=int) # shape (2368, 2)
train_y = np.array(fd['train_imagegray'],
dtype=float) / 255.0 # shape (2368, 64, 64, 1)
val_x = np.array(fd['val_locations'], dtype=float) # shape (768, 2)
val_y = np.array(fd['val_imagegray'], dtype=float) / \
255.0 # shape (768, 64, 64, 1)
image_c = train_y.shape[-1]
elif args.arch == 'conv_regressor' or args.arch == 'coordconv_regressor':
fd = h5py.File(args.data_h5, 'r')
train_y = np.array(
fd['train_normalized_locations'],
dtype=float) # shape (2368, 2)
# /255.0 # shape (2368, 64, 64, 1)
train_x = np.array(fd['train_onehots'], dtype=float)
val_y = np.array(
fd['val_normalized_locations'],
dtype=float) # shape (768, 2)
val_x = np.array(
fd['val_onehots'],
dtype=float) # shape (768, 64, 64, 1)
image_c = train_x.shape[-1]
elif args.arch == 'coordconv_rendering' or args.arch == 'deconv_bottleneck':
fd = h5py.File(args.data_h5, 'r')
train_x = np.array(fd['train_locations'], dtype=int) # shape (2368, 2)
train_y = np.array(fd['train_imagegray'],
dtype=float) / 255.0 # shape (2368, 64, 64, 1)
val_x = np.array(fd['val_locations'], dtype=float) # shape (768, 2)
val_y = np.array(fd['val_imagegray'], dtype=float) / 255.0 # shape (768, 64, 64, 1)
# add one-hot anyways to track accuracy etc. even if not used in loss
train_onehot = np.array(
fd['train_onehots'],
dtype=int) # shape (2368, 64, 64, 1)
val_onehot = np.array(
fd['val_onehots'],
dtype=int) # shape (768, 64, 64, 1)
image_c = train_y.shape[-1]
train_size = train_x.shape[0]
val_size = val_x.shape[0]
# 1. CREATE MODEL
input_coords = tf.placeholder(
shape=(None,2),
dtype='float32',
name='input_coords') # cast later in model into float
input_onehot = tf.placeholder(
shape=(None, args.x_dim, args.y_dim, 1),
dtype='float32',
name='input_onehot')
input_images = tf.placeholder(
shape=(None, args.x_dim, args.y_dim, image_c),
dtype='float32',
name='input_images')
if args.arch == 'deconv_classification':
model = DeconvPainter(l2=args.l2, x_dim=args.x_dim, y_dim=args.y_dim,
fs=args.filter_size, mul=args.channel_mul,
onthefly=DATA_GEN_ON_THE_FLY,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss)
model.a('input_coords', input_coords)
if not DATA_GEN_ON_THE_FLY:
model.a('input_onehot', input_onehot)
model([input_coords]) if DATA_GEN_ON_THE_FLY else model([input_coords, input_onehot])
if args.arch == 'conv_regressor':
regress_type = 'conv_uniform' if 'uniform' in args.data_h5 else 'conv_quarant'
model = ConvRegressor(l2=args.l2, mul=args.channel_mul,
_type=regress_type)
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
# call model on inputs
model([input_onehot, input_coords])
if args.arch == 'coordconv_regressor':
model = ConvRegressor(l2=args.l2, mul=args.channel_mul,
_type='coordconv')
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
# call model on inputs
model([input_onehot, input_coords])
if args.arch == 'conv_onehot_image':
model = ConvImagePainter(l2=args.l2, fs=args.filter_size, mul=args.channel_mul,
use_mse_loss=args.use_mse_loss, use_sigm_loss=args.use_sigm_loss,
version='working')
# version='simple') # version='simple' to hack a 9x9 all-ones filter solution
model.a('input_onehot', input_onehot)
model.a('input_images', input_images)
# call model on inputs
model([input_onehot, input_images])
if args.arch == 'deconv_rendering':
model = DeconvPainter(l2=args.l2, x_dim=args.x_dim, y_dim=args.y_dim,
fs=args.filter_size, mul=args.channel_mul,
onthefly=False,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss)
model.a('input_coords', input_coords)
model.a('input_images', input_images)
# call model on inputs
model([input_coords, input_images])
elif args.arch == 'coordconv_classification':
model = CoordConvPainter(
l2=args.l2,
x_dim=args.x_dim,
y_dim=args.y_dim,
include_r=False,
mul=args.channel_mul,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss)
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
model([input_coords, input_onehot])
#raise Exception('Not implemented yet')
elif args.arch == 'coordconv_rendering':
model = CoordConvImagePainter(
l2=args.l2,
x_dim=args.x_dim,
y_dim=args.y_dim,
include_r=False,
mul=args.channel_mul,
fs=args.filter_size,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss,
interm_loss=args.interm_loss,
no_softmax=args.no_softmax,
version='working')
# version='simple') # version='simple' to hack a 9x9 all-ones filter solution
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
model.a('input_images', input_images)
# always input three things to calculate relevant metrics
model([input_coords, input_onehot, input_images])
elif args.arch == 'deconv_bottleneck':
model = DeconvBottleneckPainter(
l2=args.l2,
x_dim=args.x_dim,
y_dim=args.y_dim,
mul=args.channel_mul,
fs=args.filter_size,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss,
interm_loss=args.interm_loss,
no_softmax=args.no_softmax,
version='working') # version='simple' to hack a 9x9 all-ones filter solution
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
model.a('input_images', input_images)
# always input three things to calculate relevant metrics
model([input_coords, input_onehot, input_images])
elif args.arch == 'upsample_conv_coords' or args.arch == 'upsample_coordconv_coords':
_coordconv = True if args.arch == 'upsample_coordconv_coords' else False
model = UpsampleConvPainter(
l2=args.l2,
x_dim=args.x_dim,
y_dim=args.y_dim,
mul=args.channel_mul,
fs=args.filter_size,
use_mse_loss=args.use_mse_loss,
use_sigm_loss=args.use_sigm_loss,
coordconv=_coordconv)
model.a('input_coords', input_coords)
model.a('input_onehot', input_onehot)
model([input_coords, input_onehot])
print 'All model weights:'
summarize_weights(model.trainable_weights)
#print 'Model summary:'
print 'Another model summary:'
model.summarize_named(prefix=' ')
print_trainable_warnings(model)
# 2. COMPUTE GRADS AND CREATE OPTIMIZER
# a placeholder for dynamic learning rate
input_lr = tf.placeholder(tf.float32, shape=[])
if args.opt == 'sgd':
opt = tf.train.MomentumOptimizer(input_lr, args.mom)
elif args.opt == 'rmsprop':
opt = tf.train.RMSPropOptimizer(input_lr, momentum=args.mom)
elif args.opt == 'adam':
opt = tf.train.AdamOptimizer(input_lr, args.beta1, args.beta2)
grads_and_vars = opt.compute_gradients(
model.loss,
model.trainable_weights,
gate_gradients=tf.train.Optimizer.GATE_GRAPH)
train_step = opt.apply_gradients(grads_and_vars)
# added to train_ and param_ collections
add_grads_and_vars_hist_summaries(grads_and_vars)
summarize_opt(opt)
print 'LR Policy:', lr_policy
# add_grad_summaries(grads_and_vars)
if not args.arch.endswith('regressor'):
image_summaries_traintest(model.logits)
if 'input_onehot' in model.named_keys():
image_summaries_traintest(model.input_onehot)
if 'input_images' in model.named_keys():
image_summaries_traintest(model.input_images)
if 'prob' in model.named_keys():
image_summaries_traintest(model.prob)
if 'center_prob' in model.named_keys():
image_summaries_traintest(model.center_prob)
if 'center_logits' in model.named_keys():
image_summaries_traintest(model.center_logits)
if 'pixelwise_prob' in model.named_keys():
image_summaries_traintest(model.pixelwise_prob)
if 'center_logits' in model.named_keys():
image_summaries_traintest(model.center_logits)
if 'sharpened_logits' in model.named_keys():
image_summaries_traintest(model.sharpened_logits)
# 3. OPTIONALLY SAVE OR LOAD VARIABLES (e.g. model params, model running
# BN means, optimization momentum, ...) and then finalize initialization
saver = tf.train.Saver(
max_to_keep=None) if (
args.output or args.load) else None
if args.load:
ckptfile, miscfile = args.load.split(':')
# Restore values directly to graph
saver.restore(sess, ckptfile)
with gzip.open(miscfile) as ff:
saved = pickle.load(ff)
buddy = saved['buddy']
else:
buddy = StatsBuddy()
buddy.tic() # call if new run OR resumed run
# Check if special layers are initialized right
#last_layer_w = [var for var in tf.global_variables() if 'painting_layer/kernel:0' in var.name][0]
#last_layer_b = [var for var in tf.global_variables() if 'painting_layer/bias:0' in var.name][0]
# Initialize any missed vars (e.g. optimization momentum, ... if not
# loaded from checkpoint)
uninitialized_vars = tf_get_uninitialized_variables(sess)
init_missed_vars = tf.variables_initializer(
uninitialized_vars, 'init_missed_vars')
sess.run(init_missed_vars)
# Print warnings about any TF vs. Keras shape mismatches
# warn_misaligned_shapes(model)
# Make sure all variables, which are model variables, have been
# initialized (e.g. model params and model running BN means)
tf_assert_all_init(sess)
# tf.global_variables_initializer().run()
# 4. SETUP TENSORBOARD LOGGING with tf.summary.merge
train_histogram_summaries = get_collection_intersection_summary(
'train_collection', 'orig_histogram')
train_scalar_summaries = get_collection_intersection_summary(
'train_collection', 'orig_scalar')
test_histogram_summaries = get_collection_intersection_summary(
'test_collection', 'orig_histogram')
test_scalar_summaries = get_collection_intersection_summary(
'test_collection', 'orig_scalar')
param_histogram_summaries = get_collection_intersection_summary(
'param_collection', 'orig_histogram')
train_image_summaries = get_collection_intersection_summary(
'train_collection', 'orig_image')
test_image_summaries = get_collection_intersection_summary(
'test_collection', 'orig_image')
writer = None
if args.output:
mkdir_p(args.output)
writer = tf.summary.FileWriter(args.output, sess.graph)
# 5. TRAIN
train_iters = (train_size) // minibatch_size + \
int(train_size % minibatch_size > 0)
if not args.skipval:
val_iters = (val_size) // minibatch_size + \
int(val_size % minibatch_size > 0)
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
while buddy.epoch < args.epochs + 1:
# How often to log data
def do_log_params(ep, it, ii): return True
def do_log_val(ep, it, ii): return True
def do_log_train(
ep,
it,
ii): return (
it < train_iters and it & it -
1 == 0 or it >= train_iters and it %
train_iters == 0) # Log on powers of two then every epoch
# 0. Log params
if args.output and do_log_params(
buddy.epoch,
buddy.train_iter,
0) and param_histogram_summaries is not None:
params_summary_str, = sess.run([param_histogram_summaries])
writer.add_summary(params_summary_str, buddy.train_iter)
# 1. Forward test on validation set
if not args.skipval:
feed_dict = {learning_phase(): 0}
if 'input_coords' in model.named_keys():
val_coords = val_y if args.arch.endswith(
'regressor') else val_x
feed_dict.update({model.input_coords: val_coords})
if 'input_onehot' in model.named_keys():
# if 'val_onehot' not in locals():
if not args.arch == 'coordconv_rendering' and not args.arch == 'deconv_bottleneck':
if args.arch == 'conv_onehot_image' or args.arch.endswith('regressor'):
val_onehot = val_x
else:
val_onehot = val_y
feed_dict.update({
model.input_onehot: val_onehot,
})
if 'input_images' in model.named_keys():
feed_dict.update({
model.input_images: val_images,
})
fetch_dict = model.trackable_dict()
if args.output and do_log_val(buddy.epoch, buddy.train_iter, 0):
if test_image_summaries is not None:
fetch_dict.update(
{'test_image_summaries': test_image_summaries})
if test_scalar_summaries is not None:
fetch_dict.update(
{'test_scalar_summaries': test_scalar_summaries})
if test_histogram_summaries is not None:
fetch_dict.update(
{'test_histogram_summaries': test_histogram_summaries})
with WithTimer('sess.run val iter', quiet=not args.verbose):
result_val = sess_run_dict(
sess, fetch_dict, feed_dict=feed_dict)
buddy.note_list(
model.trackable_names(), [
result_val[k] for k in model.trackable_names()], prefix='val_')
print (
'[%5d] [%2d/%2d] val: %s (%.3gs/i)' %
(buddy.train_iter,
buddy.epoch,
args.epochs,
buddy.epoch_mean_pretty_re(
'^val_',
style=val_style),
toc2()))
if args.output and do_log_val(buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer, buddy.train_iter, {
'mean_%s' %
name: value for name, value in buddy.epoch_mean_list_re('^val_')}, prefix='val')
if test_image_summaries is not None:
image_summary_str = result_val['test_image_summaries']
writer.add_summary(image_summary_str, buddy.train_iter)
if test_scalar_summaries is not None:
scalar_summary_str = result_val['test_scalar_summaries']
writer.add_summary(scalar_summary_str, buddy.train_iter)
if test_histogram_summaries is not None:
hist_summary_str = result_val['test_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
# 2. Possiby Snapshot, possibly quit
if args.output and args.snapshot_to and args.snapshot_every:
snap_intermed = args.snapshot_every > 0 and buddy.train_iter % args.snapshot_every == 0
#snap_end = buddy.epoch == args.epochs
snap_end = lr_policy.train_done(buddy)
if snap_intermed or snap_end:
# Snapshot network and buddy
save_path = saver.save(
sess, '%s/%s_%04d.ckpt' %
(args.output, args.snapshot_to, buddy.epoch))
print 'snappshotted model to', save_path
with gzip.open('%s/%s_misc_%04d.pkl.gz' % (args.output, args.snapshot_to, buddy.epoch), 'w') as ff:
saved = {'buddy': buddy}
pickle.dump(saved, ff)
# Snapshot evaluation data and metrics
_, _ = evaluate_net(
args, buddy, model, train_size, train_x, train_y, val_x, val_y, fd, sess)
lr = lr_policy.get_lr(buddy)
if buddy.epoch == args.epochs:
if args.ipy:
print 'Embed: at end of training (Ctrl-D to exit)'
embed()
break # Extra pass at end: just report val stats and skip training
print '********* at epoch %d, LR is %g' % (buddy.epoch, lr)
# 3. Train on training set
if args.shuffletrain:
train_order = np.random.permutation(train_size)
tic3()
for ii in xrange(train_iters):
tic2()
start_idx = ii * minibatch_size
end_idx = min(start_idx + minibatch_size, train_size)
if args.shuffletrain: # default true
batch_x = train_x[sorted(
train_order[start_idx:end_idx].tolist())]
if train_y is not None:
batch_y = train_y[sorted(
train_order[start_idx:end_idx].tolist())]
# if 'train_onehot' in locals():
if args.arch == 'coordconv_rendering' or args.arch == 'deconv_bottleneck':
batch_onehot = train_onehot[sorted(
train_order[start_idx:end_idx].tolist())]
else:
batch_x = train_x[start_idx:end_idx]
if train_y is not None:
batch_y = train_y[start_idx:end_idx]
# if 'train_onehot' in locals():
if args.arch == 'coordconv_rendering' or args.arch == 'deconv_bottleneck':
batch_onehot = train_onehot[start_idx:end_idx]
feed_dict = {learning_phase(): 1, input_lr: lr}
if 'input_coords' in model.named_keys():
batch_coords = batch_y if args.arch.endswith(
'regressor') else batch_x
feed_dict.update({model.input_coords: batch_coords})
if 'input_onehot' in model.named_keys():
# if 'batch_onehot' not in locals():
# if not (args.arch == 'coordconv_rendering' and
# args.add_interm_loss):
if not args.arch == 'coordconv_rendering' and not args.arch == 'deconv_bottleneck':
if args.arch == 'conv_onehot_image' or args.arch.endswith(
'regressor'):
batch_onehot = batch_x
else:
batch_onehot = batch_y
feed_dict.update({
model.input_onehot: batch_onehot,
})
if 'input_images' in model.named_keys():
feed_dict.update({
model.input_images: batch_images,
})
fetch_dict = model.trackable_and_update_dict()
fetch_dict.update({'train_step': train_step})
if args.output and do_log_train(buddy.epoch, buddy.train_iter, ii):
if train_histogram_summaries is not None:
fetch_dict.update(
{'train_histogram_summaries': train_histogram_summaries})
if train_scalar_summaries is not None:
fetch_dict.update(
{'train_scalar_summaries': train_scalar_summaries})
if train_image_summaries is not None:
fetch_dict.update(
{'train_image_summaries': train_image_summaries})
with WithTimer('sess.run train iter', quiet=not args.verbose):
result_train = sess_run_dict(
sess, fetch_dict, feed_dict=feed_dict)
buddy.note_weighted_list(
batch_x.shape[0], model.trackable_names(), [
result_train[k] for k in model.trackable_names()], prefix='train_')
if do_log_train(buddy.epoch, buddy.train_iter, ii):
print (
'[%5d] [%2d/%2d] train: %s (%.3gs/i)' %
(buddy.train_iter,
buddy.epoch,
args.epochs,
buddy.epoch_mean_pretty_re(
'^train_',
style=train_style),
toc2()))
if args.output and do_log_train(buddy.epoch, buddy.train_iter, ii):
if train_histogram_summaries is not None:
hist_summary_str = result_train['train_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_scalar_summaries is not None:
scalar_summary_str = result_train['train_scalar_summaries']
writer.add_summary(scalar_summary_str, buddy.train_iter)
if train_image_summaries is not None:
image_summary_str = result_train['train_image_summaries']
writer.add_summary(image_summary_str, buddy.train_iter)
log_scalars(
writer, buddy.train_iter, {
'batch_%s' %
name: value for name, value in buddy.last_list_re('^train_')}, prefix='train')
if ii > 0 and ii % 100 == 0:
print ' %d: Average iteration time over last 100 train iters: %.3gs' % (
ii, toc3() / 100)
tic3()
buddy.inc_train_iter() # after finished training a mini-batch
buddy.inc_epoch() # after finished training whole pass through set
if args.output and do_log_train(buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer, buddy.train_iter, {
'mean_%s' %
name: value for name, value in buddy.epoch_mean_list_re('^train_')}, prefix='train')
print '\nFinal'
print '%02d:%d val: %s' % (buddy.epoch,
buddy.train_iter,
buddy.epoch_mean_pretty_re(
'^val_',
style=val_style))
print '%02d:%d train: %s' % (buddy.epoch,
buddy.train_iter,
buddy.epoch_mean_pretty_re(
'^train_',
style=train_style))
print '\nEnd of training. Saving evaluation results on whole train and val set.'
final_tr_metrics, final_va_metrics = evaluate_net(
args, buddy, model, train_size, train_x, train_y, val_x, val_y, fd, sess)
print '\nFinal evaluation on whole train and val'
for name, value in final_tr_metrics.iteritems():
print 'final_stats_eval train_%s %g' % (name, value)
for name, value in final_va_metrics.iteritems():
print 'final_stats_eval val_%s %g' % (name, value)
print '\nfinal_stats epochs %g' % buddy.epoch
print 'final_stats iters %g' % buddy.train_iter
print 'final_stats time %g' % buddy.toc()
for name, value in buddy.epoch_mean_list_all():
print 'final_stats %s %g' % (name, value)
if args.output:
writer.close() # Flush and close
def main():
parser = make_standard_parser('Random Projection Experiments.',
arch_choices=arch_choices)
parser.add_argument('--vsize',
type=int,
default=100,
help='Dimension of intrinsic parmaeter space.')
parser.add_argument('--d_rate',
'--dr',
type=float,
default=0.0,
help='Dropout rate.')
parser.add_argument('--depth',
type=int,
default=2,
help='Number of layers in FNN.')
parser.add_argument('--width',
type=int,
default=100,
help='Width of layers in FNN.')
parser.add_argument('--minibatch',
'--mb',
type=int,
default=128,
help='Size of minibatch.')
parser.add_argument('--lr_ratio',
'--lrr',
type=float,
default=.1,
help='Ratio to decay LR by every LR_EPSTEP epochs.')
parser.add_argument(
'--lr_epochs',
'--lrep',
type=float,
default=0,
help='Decay LR every LR_EPSTEP epochs. 0 to turn off decay.')
parser.add_argument('--lr_steps',
'--lrst',
type=float,
default=3,
help='Max LR steps.')
parser.add_argument('--c1',
type=int,
default=6,
help='Channels in first conv layer, for LeNet.')
parser.add_argument('--c2',
type=int,
default=16,
help='Channels in second conv layer, for LeNet.')
parser.add_argument('--d1',
type=int,
default=120,
help='Channels in first dense layer, for LeNet.')
parser.add_argument('--d2',
type=int,
default=84,
help='Channels in second dense layer, for LeNet.')
parser.add_argument('--denseproj',
action='store_true',
help='Use a dense projection.')
parser.add_argument('--sparseproj',
action='store_true',
help='Use a sparse projection.')
parser.add_argument('--fastfoodproj',
action='store_true',
help='Use a fastfood projection.')
parser.add_argument('--partial_data',
'--pd',
type=float,
default=1.0,
help='Percentage of dataset.')
parser.add_argument(
'--skiptfevents',
action='store_true',
help='Skip writing tf events files even if output is used.')
args = parser.parse_args()
n_proj_specified = sum(
[args.denseproj, args.sparseproj, args.fastfoodproj])
if args.arch in arch_choices_projected:
assert n_proj_specified == 1, 'Arch "%s" requires projection. Specify exactly one of {denseproj, sparseproj, fastfoodproj} options.' % args.arch
else:
assert n_proj_specified == 0, 'Arch "%s" does not require projection, so do not specify any of {denseproj, sparseproj, fastfoodproj} options.' % args.arch
if args.denseproj:
proj_type = 'dense'
elif args.sparseproj:
proj_type = 'sparse'
else:
proj_type = 'fastfood'
train_style, val_style = ('',
'') if args.nocolor else (colorama.Fore.BLUE,
colorama.Fore.MAGENTA)
# Get a TF session registered with Keras and set numpy and TF seeds
sess = setup_session_and_seeds(args.seed)
# 0. LOAD DATA
train_h5 = h5py.File(args.train_h5, 'r')
train_x = train_h5['images']
train_y = train_h5['labels']
val_h5 = h5py.File(args.val_h5, 'r')
val_x = val_h5['images']
val_y = val_h5['labels']
# loadpath = "./dataset/ag_news.p"
# x = pickle.load(open(loadpath, "rb"))
# train_x, val_x, test_x = x[0], x[1], x[2]
# train_y, val_y, test_y = x[3], x[4], x[5]
# wordtoix, ixtoword = x[6], x[7]
# train_x = prepare_data_for_cnn(train_x, 100, 5)
# val_x = prepare_data_for_cnn(val_x, 100, 5)
# #weightInit = tf.random_uniform_initializer(-0.001, 0.001)
# #W = tf.get_variable('W', [13010, 300], initializer=weightInit)
# W = np.random.rand(13010, 300)
# #pdb.set_trace()
# train_x = [np.take(W, i, axis=0) for i in train_x]
# train_x = np.array(train_x, dtype='float32')
# val_x = [np.take(W, i, axis=0) for i in val_x]
# val_x = np.array(val_x, dtype='float32')
#pdb.set_trace()
train_x = np.array(train_x, dtype='float32')
val_x = np.array(val_x, dtype='float32')
if args.partial_data < 1.0:
n_train_ = int(train_y.size * args.partial_data)
n_test_ = int(val_y.size * args.partial_data)
train_x = train_x[:n_train_]
train_y = train_y[:n_train_]
val_x = val_x[:n_test_]
val_y = val_y[:n_test_]
# load into memory if less than 1 GB
if train_x.size * 4 + val_x.size * 4 < 1e9:
train_x, train_y = np.array(train_x), np.array(train_y)
val_x, val_y = np.array(val_x), np.array(val_y)
# 1. CREATE MODEL
randmirrors = False
randcrops = False
cropsize = None
with WithTimer('Make model'):
if args.arch == 'mnistfc_dir':
model = build_model_mnist_fc_dir(weight_decay=args.l2,
depth=args.depth,
width=args.width)
elif args.arch == 'mnistfc':
if proj_type == 'fastfood':
model = build_model_mnist_fc_fastfood(weight_decay=args.l2,
vsize=args.vsize,
depth=args.depth,
width=args.width)
else:
model = build_model_mnist_fc(weight_decay=args.l2,
vsize=args.vsize,
depth=args.depth,
width=args.width,
proj_type=proj_type)
elif args.arch == 'mnistconv':
model = build_cnn_model_mnist(weight_decay=args.l2,
vsize=args.vsize)
elif args.arch == 'mnistconv_dir':
model = build_cnn_model_direct_mnist(weight_decay=args.l2)
elif args.arch == 'cifarfc_dir':
model = build_model_cifar_fc_dir(weight_decay=args.l2,
depth=args.depth,
width=args.width)
elif args.arch == 'cifarfc':
if proj_type == 'fastfood':
model = build_model_cifar_fc_fastfood(weight_decay=args.l2,
vsize=args.vsize,
depth=args.depth,
width=args.width)
else:
model = build_model_cifar_fc(weight_decay=args.l2,
vsize=args.vsize,
depth=args.depth,
width=args.width,
proj_type=proj_type)
elif args.arch == 'mnistlenet_dir':
model = build_LeNet_direct_mnist(weight_decay=args.l2,
c1=args.c1,
c2=args.c2,
d1=args.d1,
d2=args.d2)
elif args.arch == 'mnistMLPlenet_dir':
model = build_MLPLeNet_direct_mnist(weight_decay=args.l2)
elif args.arch == 'mnistMLPlenet':
if proj_type == 'fastfood':
model = build_model_mnist_MLPLeNet_fastfood(
weight_decay=args.l2, vsize=args.vsize)
elif args.arch == 'mnistUntiedlenet_dir':
model = build_UntiedLeNet_direct_mnist(weight_decay=args.l2)
elif args.arch == 'mnistUntiedlenet':
if proj_type == 'fastfood':
model = build_model_mnist_UntiedLeNet_fastfood(
weight_decay=args.l2, vsize=args.vsize)
elif args.arch == 'cifarMLPlenet_dir':
model = build_MLPLeNet_direct_cifar(weight_decay=args.l2)
elif args.arch == 'cifarMLPlenet':
if proj_type == 'fastfood':
model = build_model_cifar_MLPLeNet_fastfood(
weight_decay=args.l2, vsize=args.vsize)
elif args.arch == 'cifarUntiedlenet_dir':
model = build_UntiedLeNet_direct_cifar(weight_decay=args.l2)
elif args.arch == 'cifarUntiedlenet':
if proj_type == 'fastfood':
model = build_model_cifar_UntiedLeNet_fastfood(
weight_decay=args.l2, vsize=args.vsize)
elif args.arch == 'mnistlenet':
if proj_type == 'fastfood':
model = build_model_mnist_LeNet_fastfood(weight_decay=args.l2,
vsize=args.vsize)
else:
model = build_LeNet_mnist(weight_decay=args.l2,
vsize=args.vsize,
proj_type=proj_type)
elif args.arch == 'cifarlenet_dir':
model = build_LeNet_direct_cifar(weight_decay=args.l2,
d_rate=args.d_rate,
c1=args.c1,
c2=args.c2,
d1=args.d1,
d2=args.d2)
elif args.arch == 'cifarlenet':
if proj_type == 'fastfood':
model = build_model_cifar_LeNet_fastfood(weight_decay=args.l2,
vsize=args.vsize,
d_rate=args.d_rate,
c1=args.c1,
c2=args.c2,
d1=args.d1,
d2=args.d2)
else:
model = build_LeNet_cifar(weight_decay=args.l2,
vsize=args.vsize,
proj_type=proj_type,
d_rate=args.d_rate)
elif args.arch == 'cifarDenseNet_dir':
model = build_DenseNet_direct_cifar(weight_decay=args.l2,
depth=25,
nb_dense_block=1,
growth_rate=12)
elif args.arch == 'cifarDenseNet':
if proj_type == 'fastfood':
model = build_DenseNet_cifar_fastfood(weight_decay=args.l2,
vsize=args.vsize,
depth=25,
nb_dense_block=1,
growth_rate=12)
elif args.arch == 'alexnet_dir':
model = build_alexnet_direct(weight_decay=args.l2,
shift_in=np.array([104, 117, 123]))
args.shuffletrain = False
randmirrors = True
randcrops = True
cropsize = (227, 227)
elif args.arch == 'squeeze_dir':
model = build_squeezenet_direct(weight_decay=args.l2,
shift_in=np.array([104, 117, 123]))
args.shuffletrain = False
randmirrors = True
randcrops = True
cropsize = (224, 224)
elif args.arch == 'alexnet':
if proj_type == 'fastfood':
model = build_alexnet_fastfood(weight_decay=args.l2,
shift_in=np.array(
[104, 117, 123]),
vsize=args.vsize)
else:
raise Exception('not implemented')
args.shuffletrain = False
randmirrors = True
randcrops = True
cropsize = (227, 227)
else:
raise Exception('Unknown network architecture: %s' % args.arch)
print 'All model weights:'
total_params = summarize_weights(model.trainable_weights)
print 'Model summary:'
model.summary()
model.print_trainable_warnings()
input_lr = tf.placeholder(tf.float32, shape=[])
lr_stepper = LRStepper(args.lr, args.lr_ratio, args.lr_epochs,
args.lr_steps)
# 2. COMPUTE GRADS AND CREATE OPTIMIZER
if args.opt == 'sgd':
opt = tf.train.MomentumOptimizer(input_lr, args.mom)
elif args.opt == 'rmsprop':
opt = tf.train.RMSPropOptimizer(input_lr, momentum=args.mom)
elif args.opt == 'adam':
opt = tf.train.AdamOptimizer(input_lr, args.beta1, args.beta2)
# Optimize w.r.t all trainable params in the model
grads_and_vars = opt.compute_gradients(
model.v.loss,
model.trainable_weights,
gate_gradients=tf.train.Optimizer.GATE_GRAPH)
train_step = opt.apply_gradients(grads_and_vars)
add_grad_summaries(grads_and_vars)
summarize_opt(opt)
# 3. OPTIONALLY SAVE OR LOAD VARIABLES (e.g. model params, model running BN means, optimization momentum, ...) and then finalize initialization
saver = tf.train.Saver(
max_to_keep=None) if (args.output or args.load) else None
if args.load:
ckptfile, miscfile = args.load.split(':')
# Restore values directly to graph
saver.restore(sess, ckptfile)
with gzip.open(miscfile) as ff:
saved = pickle.load(ff)
buddy = saved['buddy']
else:
buddy = StatsBuddy()
buddy.tic() # call if new run OR resumed run
# Initialize any missed vars (e.g. optimization momentum, ... if not loaded from checkpoint)
uninitialized_vars = tf_get_uninitialized_variables(sess)
init_missed_vars = tf.variables_initializer(uninitialized_vars,
'init_missed_vars')
sess.run(init_missed_vars)
# Print warnings about any TF vs. Keras shape mismatches
warn_misaligned_shapes(model)
# Make sure all variables, which are model variables, have been initialized (e.g. model params and model running BN means)
tf_assert_all_init(sess)
# 3.5 Normalize the overall basis matrix across the (multiple) unnormalized basis matrices for each layer
basis_matrices = []
normalizers = []
for layer in model.layers:
try:
basis_matrices.extend(layer.offset_creator.basis_matrices)
except AttributeError:
continue
try:
normalizers.extend(layer.offset_creator.basis_matrix_normalizers)
except AttributeError:
continue
if len(basis_matrices) > 0 and not args.load:
if proj_type == 'sparse':
# Norm of overall basis matrix rows (num elements in each sum == total parameters in model)
bm_row_norms = tf.sqrt(
tf.add_n([
tf.sparse_reduce_sum(tf.square(bm), 1)
for bm in basis_matrices
]))
# Assign `normalizer` Variable to these row norms to achieve normalization of the basis matrix
# in the TF computational graph
rescale_basis_matrices = [
tf.assign(var, tf.reshape(bm_row_norms, var.shape))
for var in normalizers
]
_ = sess.run(rescale_basis_matrices)
elif proj_type == 'dense':
bm_sums = [
tf.reduce_sum(tf.square(bm), 1) for bm in basis_matrices
]
divisor = tf.expand_dims(tf.sqrt(tf.add_n(bm_sums)), 1)
rescale_basis_matrices = [
tf.assign(var, var / divisor) for var in basis_matrices
]
_ = sess.run(rescale_basis_matrices)
else:
print '\nhere\n'
embed()
assert False, 'what to do with fastfood?'
# 4. SETUP TENSORBOARD LOGGING
train_histogram_summaries = get_collection_intersection_summary(
'train_collection', 'orig_histogram')
train_scalar_summaries = get_collection_intersection_summary(
'train_collection', 'orig_scalar')
val_histogram_summaries = get_collection_intersection_summary(
'val_collection', 'orig_histogram')
val_scalar_summaries = get_collection_intersection_summary(
'val_collection', 'orig_scalar')
param_histogram_summaries = get_collection_intersection_summary(
'param_collection', 'orig_histogram')
writer = None
if args.output:
mkdir_p(args.output)
if not args.skiptfevents:
writer = tf.summary.FileWriter(args.output, sess.graph)
# 5. TRAIN
train_iters = (train_y.shape[0] - 1) / args.minibatch + 1
val_iters = (val_y.shape[0] - 1) / args.minibatch + 1
impreproc = ImagePreproc()
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
fastest_avg_iter_time = 1e9
while buddy.epoch < args.epochs + 1:
# How often to log data
do_log_params = lambda ep, it, ii: False
do_log_val = lambda ep, it, ii: True
do_log_train = lambda ep, it, ii: (
it < train_iters and it & it - 1 == 0 or it >= train_iters and it %
train_iters == 0) # Log on powers of two then every epoch
# 0. Log params
if args.output and do_log_params(
buddy.epoch, buddy.train_iter, 0
) and param_histogram_summaries is not None and not args.skiptfevents:
params_summary_str, = sess.run([param_histogram_summaries])
writer.add_summary(params_summary_str, buddy.train_iter)
# 1. Evaluate val set performance
if not args.skipval:
tic2()
for ii in xrange(val_iters):
start_idx = ii * args.minibatch
batch_x = val_x[start_idx:start_idx + args.minibatch]
batch_y = val_y[start_idx:start_idx + args.minibatch]
if randcrops:
batch_x = impreproc.center_crops(batch_x, cropsize)
feed_dict = {
model.v.input_images: batch_x,
model.v.input_labels: batch_y,
K.learning_phase(): 0,
}
fetch_dict = model.trackable_dict
with WithTimer('sess.run val iter', quiet=not args.verbose):
result_val = sess_run_dict(sess,
fetch_dict,
feed_dict=feed_dict)
buddy.note_weighted_list(
batch_x.shape[0],
model.trackable_names,
[result_val[k] for k in model.trackable_names],
prefix='val_')
if args.output and not args.skiptfevents and do_log_val(
buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer,
buddy.train_iter, {
'mean_%s' % name: value
for name, value in buddy.epoch_mean_list_re('^val_')
},
prefix='buddy')
print(
'\ntime: %f. after training for %d epochs:\n%3d val: %s (%.3gs/i)'
% (buddy.toc(), buddy.epoch, buddy.train_iter,
buddy.epoch_mean_pretty_re(
'^val_', style=val_style), toc2() / val_iters))
# 2. Possiby Snapshot, possibly quit
if args.output and args.snapshot_to and args.snapshot_every:
snap_intermed = args.snapshot_every > 0 and buddy.train_iter % args.snapshot_every == 0
snap_end = buddy.epoch == args.epochs
if snap_intermed or snap_end:
# Snapshot
save_path = saver.save(
sess, '%s/%s_%04d.ckpt' %
(args.output, args.snapshot_to, buddy.epoch))
print 'snappshotted model to', save_path
with gzip.open(
'%s/%s_misc_%04d.pkl.gz' %
(args.output, args.snapshot_to, buddy.epoch), 'w') as ff:
saved = {'buddy': buddy}
pickle.dump(saved, ff)
if buddy.epoch == args.epochs:
if args.ipy:
print 'Embed: at end of training (Ctrl-D to exit)'
embed()
break # Extra pass at end: just report val stats and skip training
# 3. Train on training set
#train_order = range(train_x.shape[0])
if args.shuffletrain:
train_order = np.random.permutation(train_x.shape[0])
tic3()
for ii in xrange(train_iters):
tic2()
start_idx = ii * args.minibatch
if args.shuffletrain:
batch_x = train_x[train_order[start_idx:start_idx +
args.minibatch]]
batch_y = train_y[train_order[start_idx:start_idx +
args.minibatch]]
else:
batch_x = train_x[start_idx:start_idx + args.minibatch]
batch_y = train_y[start_idx:start_idx + args.minibatch]
if randcrops:
batch_x = impreproc.random_crops(batch_x, cropsize,
randmirrors)
feed_dict = {
model.v.input_images: batch_x,
model.v.input_labels: batch_y,
input_lr: lr_stepper.lr(buddy),
K.learning_phase(): 1,
}
fetch_dict = {'train_step': train_step}
fetch_dict.update(model.trackable_and_update_dict)
if args.output and not args.skiptfevents and do_log_train(
buddy.epoch, buddy.train_iter, ii):
if param_histogram_summaries is not None:
fetch_dict.update({
'param_histogram_summaries':
param_histogram_summaries
})
if train_histogram_summaries is not None:
fetch_dict.update({
'train_histogram_summaries':
train_histogram_summaries
})
if train_scalar_summaries is not None:
fetch_dict.update(
{'train_scalar_summaries': train_scalar_summaries})
with WithTimer('sess.run train iter', quiet=not args.verbose):
result_train = sess_run_dict(sess,
fetch_dict,
feed_dict=feed_dict)
buddy.note_weighted_list(
batch_x.shape[0],
model.trackable_names,
[result_train[k] for k in model.trackable_names],
prefix='train_')
if do_log_train(buddy.epoch, buddy.train_iter, ii):
print('%3d train: %s (%.3gs/i)' %
(buddy.train_iter,
buddy.epoch_mean_pretty_re('^train_',
style=train_style), toc2()))
if args.output and not args.skiptfevents:
if param_histogram_summaries is not None:
hist_summary_str = result_train[
'param_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_histogram_summaries is not None:
hist_summary_str = result_train[
'train_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_scalar_summaries is not None:
scalar_summary_str = result_train[
'train_scalar_summaries']
writer.add_summary(scalar_summary_str,
buddy.train_iter)
log_scalars(
writer,
buddy.train_iter, {
'batch_%s' % name: value
for name, value in buddy.last_list_re('^train_')
},
prefix='buddy')
if ii > 0 and ii % 100 == 0:
avg_iter_time = toc3() / 100
tic3()
fastest_avg_iter_time = min(fastest_avg_iter_time,
avg_iter_time)
print ' %d: Average iteration time over last 100 train iters: %.3gs' % (
ii, avg_iter_time)
buddy.inc_train_iter() # after finished training a mini-batch
buddy.inc_epoch() # after finished training whole pass through set
if args.output and not args.skiptfevents and do_log_train(
buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer,
buddy.train_iter, {
'mean_%s' % name: value
for name, value in buddy.epoch_mean_list_re('^train_')
},
prefix='buddy')
print '\nFinal'
print '%02d:%d val: %s' % (buddy.epoch, buddy.train_iter,
buddy.epoch_mean_pretty_re('^val_',
style=val_style))
print '%02d:%d train: %s' % (buddy.epoch, buddy.train_iter,
buddy.epoch_mean_pretty_re('^train_',
style=train_style))
print '\nfinal_stats epochs %g' % buddy.epoch
print 'final_stats iters %g' % buddy.train_iter
print 'final_stats time %g' % buddy.toc()
print 'final_stats total_params %g' % total_params
print 'final_stats fastest_avg_iter_time %g' % fastest_avg_iter_time
for name, value in buddy.epoch_mean_list_all():
print 'final_stats %s %g' % (name, value)
if args.output and not args.skiptfevents:
writer.close() # Flush and close
def main():
lr_policy_choices = ('constant', 'step', 'valstep')
parser = make_standard_parser('Region Proposal Net',
arch_choices=arch_choices,
skip_train=True,
skip_val=True)
parser.add_argument(
'--num',
'-N',
type=int,
default=2,
help='Load the Field-of-MNIST dataset with NUM digits per image.')
parser.add_argument('--lrpolicy',
type=str,
default='constant',
choices=lr_policy_choices,
help='LR policy.')
parser.add_argument('--lrstepratio',
type=float,
default=.1,
help='LR policy step ratio.')
parser.add_argument('--lrmaxsteps',
type=int,
default=5,
help='LR policy step ratio.')
parser.add_argument('--lrstepevery',
type=int,
default=50,
help='LR policy step ratio.')
parser.add_argument('--clip',
action='store_true',
help='clip predicted and ground truth boxes.')
parser.add_argument('--same',
action='store_true',
help='Use `same` filter instead of `valid` in conv.')
parser.add_argument('--showbox',
action='store_true',
help='show moved box during training.')
args = parser.parse_args()
if args.lrpolicy == 'step':
lr_policy = LRPolicyStep(args)
elif args.lrpolicy == 'valstep':
lr_policy = LRPolicyValStep(args)
else:
lr_policy = LRPolicyConstant(args)
minibatch_size = 1
train_style, val_style = ('',
'') if args.nocolor else (colorama.Fore.BLUE,
colorama.Fore.MAGENTA)
sess = setup_session_and_seeds(args.seed, assert_gpu=not args.cpu)
# 0. Load data
#train_ims, train_pos, train_class, valid_ims, valid_pos, valid_class, _, _, _ = load_tvt_n_per_field(args.num)
#train_ims = train_ims[:5000] # (5000, 64, 64, 1)
#train_pos = train_pos[:5000] # (5000, 2, 4)
#valid_ims = valid_ims[:1000]
#valid_pos = valid_pos[:1000]
#_ims, _pos, _class, _, _, _, _, _, _ = load_tvt_n_per_field_centercrop(args.num)
ff = h5py.File('data/field_of_mnist_cropped_64x64_5objs.h5', 'r')
train_ims = np.array(ff['train_ims']) # (9000, 64, 64, 1)
train_pos = np.array(
ff['train_pos']) # (9000, 5, 4), parts of boxes may be out of canvas
train_class = np.array(ff['train_class']) # (9000, 5)
valid_ims = np.array(ff['valid_ims']) # (1000, 64, 64, 1)
valid_pos = np.array(
ff['valid_pos']) # (1000, 5, 4), parts of boxes may be out of canvas
valid_class = np.array(ff['valid_class']) # (1000, 5)
ff.close()
im_h, im_w, im_c = train_ims.shape[1], train_ims.shape[2], train_ims.shape[
3]
train_size = train_ims.shape[0]
val_size = valid_ims.shape[0]
print(('Data loaded:\n\timage shape: {}x{}x{}'.format(im_h, im_w, im_c)))
print(('\ttrain size: {}\n\ttest size: {}'.format(train_size, val_size)))
print(('\tnumber of objects per image: {}'.format(train_pos.shape[1])))
####################
# RPN prameters
####################
rpn_params = RPNParams(anchors=np.array([(15, 15), (20, 20), (25, 25),
(15, 20), (20, 25), (20, 15),
(25, 20), (15, 25), (25, 15)]),
rpn_hidden_dim=32,
zero_box_conv=False,
weight_init_std=0.01,
anchor_scale=1.0)
bsamp_params = BoxSamplerParams(hi_thresh=0.5,
lo_thresh=0.1,
sample_size=12)
nms_params = NMSParams(
nms_thresh=0.8,
max_proposals=10,
)
# 1. CREATE MODEL
input_images = tf.placeholder(shape=(None, im_h, im_w, im_c),
dtype='float32',
name='input_images')
input_gtbox = tf.placeholder(shape=(train_pos.shape[1], 4),
dtype='float32',
name='input_gtbox')
if args.arch == 'rpn_sampler':
model = RegionProposalSampler(rpn_params,
bsamp_params,
nms_params,
l2=args.l2,
im_h=im_h,
im_w=im_w,
coordconv=False,
clip=args.clip,
filtersame=args.same)
elif args.arch == 'coord_rpn_sampler':
model = RegionProposalSampler(rpn_params,
bsamp_params,
nms_params,
l2=args.l2,
im_h=im_h,
im_w=im_w,
coordconv=True,
clip=args.clip,
filtersame=args.same)
else:
raise ValueError('Architecture {} unknown'.format(args.arch))
if args.same:
anchors = make_anchors_mnist_same(
(16, 16), minibatch_size,
rpn_params.anchors) # (batch, 16, 16, 4k)
input_anchors = tf.placeholder(shape=(16, 16,
4 * rpn_params.num_anchors),
dtype='float32',
name='input_anchors')
else:
anchors = make_anchors_mnist((13, 13), minibatch_size,
rpn_params.anchors) # (batch, 13, 13, 4k)
input_anchors = tf.placeholder(shape=(13, 13,
4 * rpn_params.num_anchors),
dtype='float32',
name='input_anchors')
anchors = anchors[0]
model.a('input_images', input_images)
model.a('input_anchors', input_anchors)
model.a('input_gtbox', input_gtbox)
model([input_images, input_anchors, input_gtbox])
print('All model weights:')
summarize_weights(model.trainable_weights)
#print 'Model summary:'
print('Another model summary:')
model.summarize_named(prefix=' ')
print_trainable_warnings(model)
# 2. COMPUTE GRADS AND CREATE OPTIMIZER
input_lr = tf.placeholder(
tf.float32, shape=[]) # a placeholder for dynamic learning rate
if args.opt == 'sgd':
opt = tf.train.MomentumOptimizer(input_lr, args.mom)
elif args.opt == 'rmsprop':
opt = tf.train.RMSPropOptimizer(input_lr, momentum=args.mom)
elif args.opt == 'adam':
opt = tf.train.AdamOptimizer(input_lr, args.beta1, args.beta2)
grads_and_vars = opt.compute_gradients(
model.loss,
model.trainable_weights,
gate_gradients=tf.train.Optimizer.GATE_GRAPH)
train_step = opt.apply_gradients(grads_and_vars)
add_grads_and_vars_hist_summaries(
grads_and_vars) # added to train_ and param_ collections
summarize_opt(opt)
print(('LR Policy:', lr_policy))
# 3. OPTIONALLY SAVE OR LOAD VARIABLES (e.g. model params, model running BN means, optimization momentum, ...) and then finalize initialization
saver = tf.train.Saver(
max_to_keep=None) if (args.output or args.load) else None
if args.load:
ckptfile, miscfile = args.load.split(':')
# Restore values directly to graph
saver.restore(sess, ckptfile)
with gzip.open(miscfile) as ff:
saved = pickle.load(ff)
buddy = saved['buddy']
else:
buddy = StatsBuddy()
buddy.tic() # call if new run OR resumed run
# Check if special layers are initialized right
#last_layer_w = [var for var in tf.global_variables() if 'painting_layer/kernel:0' in var.name][0]
#last_layer_b = [var for var in tf.global_variables() if 'painting_layer/bias:0' in var.name][0]
# Initialize any missed vars (e.g. optimization momentum, ... if not loaded from checkpoint)
uninitialized_vars = tf_get_uninitialized_variables(sess)
init_missed_vars = tf.variables_initializer(uninitialized_vars,
'init_missed_vars')
sess.run(init_missed_vars)
tf_assert_all_init(sess)
# 4. SETUP TENSORBOARD LOGGING with tf.summary.merge
train_histogram_summaries = get_collection_intersection_summary(
'train_collection', 'orig_histogram')
train_scalar_summaries = get_collection_intersection_summary(
'train_collection', 'orig_scalar')
test_histogram_summaries = get_collection_intersection_summary(
'test_collection', 'orig_histogram')
test_scalar_summaries = get_collection_intersection_summary(
'test_collection', 'orig_scalar')
param_histogram_summaries = get_collection_intersection_summary(
'param_collection', 'orig_histogram')
train_image_summaries = get_collection_intersection_summary(
'train_collection', 'orig_image')
test_image_summaries = get_collection_intersection_summary(
'test_collection', 'orig_image')
writer = None
if args.output:
mkdir_p(args.output)
writer = tf.summary.FileWriter(args.output, sess.graph)
# 5. TRAIN
train_iters = (train_size) // minibatch_size
if not args.skipval:
val_iters = (val_size) // minibatch_size
if args.output:
show_indices = np.random.permutation(val_size)[:9]
mkdir_p('{}/figures'.format(args.output))
if args.ipy:
print('Embed: before train / val loop (Ctrl-D to continue)')
embed()
while buddy.epoch < args.epochs + 1:
# How often to log data
do_log_params = lambda ep, it, ii: True
do_log_val = lambda ep, it, ii: True
do_log_train = lambda ep, it, ii: (
it < train_iters and it & it - 1 == 0 or it >= train_iters and it %
train_iters == 0) # Log on powers of two then every epoch
# 0. Log params
if args.output and do_log_params(
buddy.epoch, buddy.train_iter,
0) and param_histogram_summaries is not None:
params_summary_str, = sess.run([param_histogram_summaries])
writer.add_summary(params_summary_str, buddy.train_iter)
# 1. Forward test on validation set
if not args.skipval:
for ii in range(val_iters):
tic2()
start_idx = ii * minibatch_size
end_idx = min(start_idx + minibatch_size, val_size)
if not end_idx > start_idx:
continue
feed_dict = {
model.input_images: valid_ims[start_idx:end_idx],
model.input_anchors: anchors,
model.input_gtbox: valid_pos[start_idx:end_idx][0],
learning_phase(): 0
}
fetch_dict = model.trackable_dict()
if args.output and do_log_val(buddy.epoch, buddy.train_iter,
0):
if test_image_summaries is not None:
fetch_dict.update(
{'test_image_summaries': test_image_summaries})
if test_scalar_summaries is not None:
fetch_dict.update(
{'test_scalar_summaries': test_scalar_summaries})
if test_histogram_summaries is not None:
fetch_dict.update({
'test_histogram_summaries':
test_histogram_summaries
})
with WithTimer('sess.run val iter', quiet=not args.verbose):
result_val = sess_run_dict(sess,
fetch_dict,
feed_dict=feed_dict)
## DEBUG
## dynamic p_size and n_size, shouldn slightly very every sample
#if ii > 0 and ii % 100 == 0:
# print 'VALIDATION --- '
# print sess.run(model.p_size, feed_dict=feed_dict)
# print sess.run(model.n_size, feed_dict=feed_dict)
## END DEBUG
buddy.note_weighted_list(
minibatch_size,
model.trackable_names(),
[result_val[k] for k in model.trackable_names()],
prefix='val_')
# Done all val set
print(('[%5d] [%2d/%2d] val: %s (%.3gs/i)' %
(buddy.train_iter, buddy.epoch, args.epochs,
buddy.epoch_mean_pretty_re('^val_',
style=val_style), toc2())))
if args.output and do_log_val(buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer,
buddy.train_iter, {
'mean_%s' % name: value
for name, value in buddy.epoch_mean_list_re('^val_')
},
prefix='val')
if test_image_summaries is not None:
image_summary_str = result_val['test_image_summaries']
writer.add_summary(image_summary_str, buddy.train_iter)
if test_scalar_summaries is not None:
scalar_summary_str = result_val['test_scalar_summaries']
writer.add_summary(scalar_summary_str, buddy.train_iter)
if test_histogram_summaries is not None:
hist_summary_str = result_val['test_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
# show some boxes
if args.showbox: #and (valid_losses[epoch]/previous_best < 1- args.thresh):
show_indices = [55, 555, 678]
for show_idx in show_indices:
[pos_box, pos_score, neg_box, neg_score] = sess.run(
[
model.pos_box, model.pos_score, model.neg_box,
model.neg_score
],
feed_dict={
model.input_images: valid_ims[show_idx:show_idx + 1],
model.input_anchors: anchors,
model.input_gtbox: valid_pos[show_idx],
learning_phase(): 0
})
subplot(1, 3, show_indices.index(show_idx) + 1)
#plot_boxes_pos_neg(valid_ims[show_idx], valid_pos[show_idx], pos_box, neg_box)
plot_pos_boxes(valid_ims[show_idx],
valid_pos[show_idx],
pos_box,
pos_score,
showlabel=False)
show()
if args.output:
switch_backend('Agg')
plot_fetch_dict = {
'pos_box': model.pos_box,
'pos_score': model.pos_score,
'neg_box': model.neg_box,
'neg_score': model.neg_score,
'nms_boxes': model.nms_boxes,
'nms_scores': model.nms_scores,
}
#fig1, ax1 = subplots(3,3) # plot train boxes
#fig2, ax2 = subplots(3,3) # plot test/nms boxes
for cc, show_idx in enumerate(show_indices, 1):
feed_dict = {
model.input_images: valid_ims[show_idx:show_idx + 1],
model.input_anchors: anchors,
model.input_gtbox: valid_pos[show_idx],
learning_phase(): 0
}
result_plots = sess_run_dict(sess,
plot_fetch_dict,
feed_dict=feed_dict)
fig1 = figure(1)
subplot(3, 3, cc)
plot_boxes_pos_neg(valid_ims[show_idx], valid_pos[show_idx],
result_plots['pos_box'],
result_plots['neg_box'])
fig2 = figure(2)
subplot(3, 3, cc)
#plot_pos_boxes(valid_ims[show_idx], valid_pos[show_idx], result_plots['nms_boxes'], result_plots['nms_scores'], showlabel=False)
# normalize scores between 0 and 5, to be used as line width
_score_as_lw = 5 * (result_plots['nms_scores'] -
result_plots['nms_scores'].min()) / (
result_plots['nms_scores'].max() -
result_plots['nms_scores'].min())
plot_pos_boxes_thickness(valid_ims[show_idx],
valid_pos[show_idx],
result_plots['nms_boxes'],
result_plots['nms_scores'])
fig1.set_size_inches(10, 10)
fig1.savefig('{}/figures/pos_neg_train_box_epoch_{}.png'.format(
args.output, buddy.epoch),
dpi=100)
fig2.set_size_inches(10, 10)
fig2.savefig('{}/figures/nms_test_box_epoch_{}.png'.format(
args.output, buddy.epoch),
dpi=100)
# plot test/nms boxes
fig, _ = subplots()
# 2. Possiby Snapshot, possibly quit
if args.output and args.snapshot_to and args.snapshot_every:
snap_intermed = args.snapshot_every > 0 and buddy.train_iter % args.snapshot_every == 0
#snap_end = buddy.epoch == args.epochs
snap_end = lr_policy.train_done(buddy)
if snap_intermed or snap_end:
# Snapshot network and buddy
save_path = saver.save(
sess, '%s/%s_%04d.ckpt' %
(args.output, args.snapshot_to, buddy.epoch))
print(('snappshotted model to', save_path))
with gzip.open(
'%s/%s_misc_%04d.pkl.gz' %
(args.output, args.snapshot_to, buddy.epoch), 'w') as ff:
saved = {'buddy': buddy}
pickle.dump(saved, ff)
lr = lr_policy.get_lr(buddy)
if buddy.epoch == args.epochs:
if args.ipy:
print('Embed: at end of training (Ctrl-D to exit)')
embed()
break # Extra pass at end: just report val stats and skip training
print(('********* at epoch %d, LR is %g' % (buddy.epoch, lr)))
# 3. Train on training set
if args.shuffletrain:
train_order = np.random.permutation(train_size)
tic3()
for ii in range(train_iters):
tic2()
start_idx = ii * minibatch_size
end_idx = min(start_idx + minibatch_size, train_size)
if not end_idx > start_idx:
continue
if args.shuffletrain: # default true
batch_ims = train_ims[sorted(
train_order[start_idx:end_idx].tolist())]
batch_pos = train_pos[sorted(
train_order[start_idx:end_idx].tolist())]
else:
batch_ims = train_ims[start_idx:end_idx]
batch_pos = train_pos[start_idx:end_idx]
feed_dict = {
model.input_images: batch_ims,
model.input_anchors: anchors,
model.input_gtbox: batch_pos[0],
learning_phase(): 1,
input_lr: lr
}
fetch_dict = model.trackable_and_update_dict()
fetch_dict.update({'train_step': train_step})
if args.output and do_log_train(buddy.epoch, buddy.train_iter, ii):
if train_histogram_summaries is not None:
fetch_dict.update({
'train_histogram_summaries':
train_histogram_summaries
})
if train_scalar_summaries is not None:
fetch_dict.update(
{'train_scalar_summaries': train_scalar_summaries})
if train_image_summaries is not None:
fetch_dict.update(
{'train_image_summaries': train_image_summaries})
with WithTimer('sess.run train iter', quiet=not args.verbose):
result_train = sess_run_dict(sess,
fetch_dict,
feed_dict=feed_dict)
buddy.note_weighted_list(
minibatch_size,
model.trackable_names(),
[result_train[k] for k in model.trackable_names()],
prefix='train_')
if do_log_train(buddy.epoch, buddy.train_iter, ii):
print(('[%5d] [%2d/%2d] train: %s (%.3gs/i)' %
(buddy.train_iter, buddy.epoch, args.epochs,
buddy.epoch_mean_pretty_re(
'^train_', style=train_style), toc2())))
if args.output and do_log_train(buddy.epoch, buddy.train_iter, ii):
if train_histogram_summaries is not None:
hist_summary_str = result_train[
'train_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_scalar_summaries is not None:
scalar_summary_str = result_train['train_scalar_summaries']
writer.add_summary(scalar_summary_str, buddy.train_iter)
if train_image_summaries is not None:
image_summary_str = result_train['train_image_summaries']
writer.add_summary(image_summary_str, buddy.train_iter)
log_scalars(
writer,
buddy.train_iter, {
'batch_%s' % name: value
for name, value in buddy.last_list_re('^train_')
},
prefix='train')
if ii > 0 and ii % 100 == 0:
print((
' %d: Average iteration time over last 100 train iters: %.3gs'
% (ii, toc3() / 100)))
tic3()
buddy.inc_train_iter() # after finished training a mini-batch
buddy.inc_epoch() # after finished training whole pass through set
if args.output and do_log_train(buddy.epoch, buddy.train_iter, 0):
log_scalars(
writer,
buddy.train_iter, {
'mean_%s' % name: value
for name, value in buddy.epoch_mean_list_re('^train_')
},
prefix='train')
print('\nFinal')
print(('%02d:%d val: %s' %
(buddy.epoch, buddy.train_iter,
buddy.epoch_mean_pretty_re('^val_', style=val_style))))
print(('%02d:%d train: %s' %
(buddy.epoch, buddy.train_iter,
buddy.epoch_mean_pretty_re('^train_', style=train_style))))
print(
'\nEnd of training. Saving evaluation results on whole train and val set.'
)
if args.output:
writer.close() # Flush and close
def main():
parser = make_standard_parser('Low Rank Basis experiments.', skip_train=True, skip_val=True, arch_choices=['one'])
parser.add_argument('--DD', type=int, default=1000, help='Dimension of full parameter space.')
parser.add_argument('--vsize', type=int, default=100, help='Dimension of intrinsic parameter space.')
parser.add_argument('--lr_ratio', '--lrr', type=float, default=.5, help='Ratio to decay LR by every LR_EPSTEP epochs.')
parser.add_argument('--lr_epochs', '--lrep', type=float, default=0, help='Decay LR every LR_EPSTEP epochs. 0 to turn off decay.')
parser.add_argument('--lr_steps', '--lrst', type=float, default=3, help='Max LR steps.')
parser.add_argument('--denseproj', action='store_true', help='Use a dense projection.')
parser.add_argument('--skiptfevents', action='store_true', help='Skip writing tf events files even if output is used.')
args = parser.parse_args()
if args.denseproj:
proj_type = 'dense'
else:
proj_type = None
train_style, val_style = ('', '') if args.nocolor else (colorama.Fore.BLUE, colorama.Fore.MAGENTA)
# Get a TF session registered with Keras and set numpy and TF seeds
sess = setup_session_and_seeds(args.seed)
# 1. CREATE MODEL
with WithTimer('Make model'):
if args.denseproj:
model = build_toy(weight_decay=args.l2, DD=args.DD, groups=10, vsize=args.vsize, proj=True)
else:
model = build_toy(weight_decay=args.l2, DD=args.DD, proj=False)
print 'All model weights:'
total_params = summarize_weights(model.trainable_weights)
print 'Model summary:'
model.summary()
model.print_trainable_warnings()
input_lr = tf.placeholder(tf.float32, shape=[])
lr_stepper = LRStepper(args.lr, args.lr_ratio, args.lr_epochs, args.lr_steps)
# 2. COMPUTE GRADS AND CREATE OPTIMIZER
if args.opt == 'sgd':
opt = tf.train.MomentumOptimizer(input_lr, args.mom)
elif args.opt == 'rmsprop':
opt = tf.train.RMSPropOptimizer(input_lr, momentum=args.mom)
elif args.opt == 'adam':
opt = tf.train.AdamOptimizer(input_lr, args.beta1, args.beta2)
# Optimize w.r.t all trainable params in the model
grads_and_vars = opt.compute_gradients(model.v.loss, model.trainable_weights, gate_gradients=tf.train.Optimizer.GATE_GRAPH)
train_step = opt.apply_gradients(grads_and_vars)
add_grad_summaries(grads_and_vars)
summarize_opt(opt)
# 3. OPTIONALLY SAVE OR LOAD VARIABLES (e.g. model params, model running BN means, optimization momentum, ...) and then finalize initialization
saver = tf.train.Saver(max_to_keep=None) if (args.output or args.load) else None
if args.load:
ckptfile, miscfile = args.load.split(':')
# Restore values directly to graph
saver.restore(sess, ckptfile)
with gzip.open(miscfile) as ff:
saved = pickle.load(ff)
buddy = saved['buddy']
else:
buddy = StatsBuddy()
buddy.tic() # call if new run OR resumed run
# Initialize any missed vars (e.g. optimization momentum, ... if not loaded from checkpoint)
uninitialized_vars = tf_get_uninitialized_variables(sess)
init_missed_vars = tf.variables_initializer(uninitialized_vars, 'init_missed_vars')
sess.run(init_missed_vars)
# Print warnings about any TF vs. Keras shape mismatches
warn_misaligned_shapes(model)
# Make sure all variables, which are model variables, have been initialized (e.g. model params and model running BN means)
tf_assert_all_init(sess)
# Choose between sparsified and dense projection matrix if using them
#SparseRM = True
# 3.5 Normalize the overall basis matrix across the (multiple) unnormalized basis matrices for each layer
basis_matrices = []
normalizers = []
for layer in model.layers:
try:
basis_matrices.extend(layer.offset_creator.basis_matrices)
except AttributeError:
continue
try:
normalizers.extend(layer.offset_creator.basis_matrix_normalizers)
except AttributeError:
continue
if len(basis_matrices) > 0 and not args.load:
if proj_type == 'sparse':
# Norm of overall basis matrix rows (num elements in each sum == total parameters in model)
bm_row_norms = tf.sqrt(tf.add_n([tf.sparse_reduce_sum(tf.square(bm), 1) for bm in basis_matrices]))
# Assign `normalizer` Variable to these row norms to achieve normalization of the basis matrix
# in the TF computational graph
rescale_basis_matrices = [tf.assign(var, tf.reshape(bm_row_norms,var.shape)) for var in normalizers]
_ = sess.run(rescale_basis_matrices)
elif proj_type == 'dense':
bm_sums = [tf.reduce_sum(tf.square(bm), 1) for bm in basis_matrices]
divisor = tf.expand_dims(tf.sqrt(tf.add_n(bm_sums)), 1)
rescale_basis_matrices = [tf.assign(var, var / divisor) for var in basis_matrices]
sess.run(rescale_basis_matrices)
else:
print '\nhere\n'
embed()
assert False, 'what to do with fastfood?'
# 4. SETUP TENSORBOARD LOGGING
train_histogram_summaries = get_collection_intersection_summary('train_collection', 'orig_histogram')
train_scalar_summaries = get_collection_intersection_summary('train_collection', 'orig_scalar')
val_histogram_summaries = get_collection_intersection_summary('val_collection', 'orig_histogram')
val_scalar_summaries = get_collection_intersection_summary('val_collection', 'orig_scalar')
param_histogram_summaries = get_collection_intersection_summary('param_collection', 'orig_histogram')
writer = None
if args.output:
mkdir_p(args.output)
if not args.skiptfevents:
writer = tf.summary.FileWriter(args.output, sess.graph)
# 5. TRAIN
train_iters = 1
val_iters = 1
if args.ipy:
print 'Embed: before train / val loop (Ctrl-D to continue)'
embed()
fastest_avg_iter_time = 1e9
while buddy.epoch < args.epochs + 1:
# How often to log data
do_log_params = lambda ep, it, ii: False
do_log_val = lambda ep, it, ii: True
do_log_train = lambda ep, it, ii: (it < train_iters and it & it-1 == 0 or it>=train_iters and it%train_iters == 0) # Log on powers of two then every epoch
# 0. Log params
if args.output and do_log_params(buddy.epoch, buddy.train_iter, 0) and param_histogram_summaries is not None and not args.skiptfevents:
params_summary_str, = sess.run([param_histogram_summaries])
writer.add_summary(params_summary_str, buddy.train_iter)
# 1. Evaluate val set performance
if not args.skipval:
tic2()
for ii in xrange(val_iters):
with WithTimer('val iter %d/%d'%(ii, val_iters), quiet=not args.verbose):
feed_dict = {
K.learning_phase(): 0,
}
fetch_dict = model.trackable_dict
with WithTimer('sess.run val iter', quiet=not args.verbose):
result_val = sess_run_dict(sess, fetch_dict, feed_dict=feed_dict)
buddy.note_weighted_list(1, model.trackable_names, [result_val[k] for k in model.trackable_names], prefix='val_')
if args.output and not args.skiptfevents and do_log_val(buddy.epoch, buddy.train_iter, 0):
log_scalars(writer, buddy.train_iter,
{'mean_%s' % name: value for name, value in buddy.epoch_mean_list_re('^val_')},
prefix='buddy')
print ('\ntime: %f. after training for %d epochs:\n%3d val: %s (%.3gs/i)'
% (buddy.toc(), buddy.epoch, buddy.train_iter, buddy.epoch_mean_pretty_re('^val_', style=val_style), toc2() / val_iters))
# 2. Possiby Snapshot, possibly quit
if args.output and args.snapshot_to and args.snapshot_every:
snap_intermed = args.snapshot_every > 0 and buddy.train_iter % args.snapshot_every == 0
snap_end = buddy.epoch == args.epochs
if snap_intermed or snap_end:
# Snapshot
save_path = saver.save(sess, '%s/%s_%04d.ckpt' % (args.output, args.snapshot_to, buddy.epoch))
print 'snappshotted model to', save_path
with gzip.open('%s/%s_misc_%04d.pkl.gz' % (args.output, args.snapshot_to, buddy.epoch), 'w') as ff:
saved = {'buddy': buddy}
pickle.dump(saved, ff)
if buddy.epoch == args.epochs:
if args.ipy:
print 'Embed: at end of training (Ctrl-D to exit)'
embed()
break # Extra pass at end: just report val stats and skip training
# 3. Train on training set
#train_order = range(train_x.shape[0])
tic3()
for ii in xrange(train_iters):
with WithTimer('train iter %d/%d'%(ii, train_iters), quiet=not args.verbose):
tic2()
feed_dict = {
input_lr: lr_stepper.lr(buddy),
K.learning_phase(): 1,
}
fetch_dict = {'train_step': train_step}
fetch_dict.update(model.trackable_and_update_dict)
if args.output and not args.skiptfevents and do_log_train(buddy.epoch, buddy.train_iter, ii):
if param_histogram_summaries is not None:
fetch_dict.update({'param_histogram_summaries': param_histogram_summaries})
if train_histogram_summaries is not None:
fetch_dict.update({'train_histogram_summaries': train_histogram_summaries})
if train_scalar_summaries is not None:
fetch_dict.update({'train_scalar_summaries': train_scalar_summaries})
with WithTimer('sess.run train iter', quiet=not args.verbose):
result_train = sess_run_dict(sess, fetch_dict, feed_dict=feed_dict)
buddy.note_weighted_list(1, model.trackable_names, [result_train[k] for k in model.trackable_names], prefix='train_')
if do_log_train(buddy.epoch, buddy.train_iter, ii):
print ('%3d train: %s (%.3gs/i)' % (buddy.train_iter, buddy.epoch_mean_pretty_re('^train_', style=train_style), toc2()))
if args.output and not args.skiptfevents:
if param_histogram_summaries is not None:
hist_summary_str = result_train['param_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_histogram_summaries is not None:
hist_summary_str = result_train['train_histogram_summaries']
writer.add_summary(hist_summary_str, buddy.train_iter)
if train_scalar_summaries is not None:
scalar_summary_str = result_train['train_scalar_summaries']
writer.add_summary(scalar_summary_str, buddy.train_iter)
log_scalars(writer, buddy.train_iter,
{'batch_%s' % name: value for name, value in buddy.last_list_re('^train_')},
prefix='buddy')
log_scalars(writer, buddy.train_iter, {'batch_lr': lr_stepper.lr(buddy)}, prefix='buddy')
if ii > 0 and ii % 100 == 0:
avg_iter_time = toc3() / 100; tic3()
fastest_avg_iter_time = min(fastest_avg_iter_time, avg_iter_time)
print ' %d: Average iteration time over last 100 train iters: %.3gs' % (ii, avg_iter_time)
buddy.inc_train_iter() # after finished training a mini-batch
buddy.inc_epoch() # after finished training whole pass through set
if args.output and not args.skiptfevents and do_log_train(buddy.epoch, buddy.train_iter, 0):
log_scalars(writer, buddy.train_iter,
{'mean_%s' % name: value for name,value in buddy.epoch_mean_list_re('^train_')},
prefix='buddy')
print '\nFinal'
print '%02d:%d val: %s' % (buddy.epoch, buddy.train_iter, buddy.epoch_mean_pretty_re('^val_', style=val_style))
print '%02d:%d train: %s' % (buddy.epoch, buddy.train_iter, buddy.epoch_mean_pretty_re('^train_', style=train_style))
print '\nfinal_stats epochs %g' % buddy.epoch
print 'final_stats iters %g' % buddy.train_iter
print 'final_stats time %g' % buddy.toc()
print 'final_stats total_params %g' % total_params
print 'final_stats fastest_avg_iter_time %g' % fastest_avg_iter_time
for name, value in buddy.epoch_mean_list_all():
print 'final_stats %s %g' % (name, value)
if args.output and not args.skiptfevents:
writer.close() # Flush and close