strides=dict(str_h=2, str_w=4)),
Pooling(2, strides=2),
Conv((5, 5, 64),
init=init,
batch_norm=True,
activation=Rectlin(),
strides=dict(str_h=1, str_w=2)),
DeepBiRNN(128,
init=GlorotUniform(),
batch_norm=True,
activation=Rectlin(),
reset_cells=True,
depth=3),
RecurrentMean(),
Affine(nout=common['nclasses'], init=init, activation=Softmax())
]
model = Model(layers=layers)
opt = Adagrad(learning_rate=0.01)
metric = Misclassification()
callbacks = Callbacks(model, eval_set=val, metric=metric, **args.callback_args)
cost = GeneralizedCost(costfunc=CrossEntropyMulti())
model.fit(train,
optimizer=opt,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
print('Misclassification error = %.1f%%' %
(model.eval(val, metric=metric) * 100))
layers = [
Affine(nout=50, init=w, bias=b, activation=Rectlin()),
Dropout(keep=0.5),
Affine(nout=50, init=w, bias=b, activation=Rectlin()),
Dropout(keep=0.4),
Affine(nout=3, init=w, bias=b, activation=Softmax()),
Dropout(keep=0.3)
]
# Optimizer
optimizer = GradientDescentMomentum(0.1,
momentum_coef=0.9,
stochastic_round=args.rounding)
# Cost
cost = GeneralizedCost(costfunc=MeanSquared())
model = Model(layers=layers)
callbacks = Callbacks(model, eval_set=val_iter, **args.callback_args)
# Training
model.fit(train_iter,
optimizer=optimizer,
num_epochs=1,
cost=cost,
callbacks=callbacks)
# Evluate
evaluate(model, val_iter, Metric=Misclassification())
reset_cells=False),
Affine(train_set.nfeatures, init, bias=init, activation=Identity())
]
else:
layers = [
LSTM(hidden,
init,
activation=Logistic(),
gate_activation=Tanh(),
reset_cells=True),
RecurrentLast(),
Affine(train_set.nfeatures, init, bias=init, activation=Identity())
]
model = Model(layers=layers)
cost = GeneralizedCost(MeanSquared())
optimizer = RMSProp(stochastic_round=args.rounding)
callbacks = Callbacks(model, eval_set=valid_set, **args.callback_args)
# fit model
model.fit(train_set,
optimizer=optimizer,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
# =======visualize how the model does on validation set==============
# run the trained model on train and valid dataset and see how the outputs
# match
train_output = model.get_outputs(train_set).reshape(
def main():
# parse the command line arguments
parser = NeonArgparser(__doc__)
args = parser.parse_args()
logger = logging.getLogger()
logger.setLevel(args.log_thresh)
#Set up batch iterator for training images
print "Setting up data batch loaders..."
train = ImgMaster(repo_dir='dataTmp',
set_name='train',
inner_size=120,
subset_pct=100)
val = ImgMaster(repo_dir='dataTmp',
set_name='train',
inner_size=120,
subset_pct=100,
do_transforms=False)
test = ImgMaster(repo_dir='dataTestTmp',
set_name='train',
inner_size=120,
subset_pct=100,
do_transforms=False)
train.init_batch_provider()
val.init_batch_provider()
test.init_batch_provider()
print "Constructing network..."
#Create AlexNet architecture
model = constuct_network()
#model.load_weights(args.model_file)
# drop weights LR by 1/250**(1/3) at epochs (23, 45, 66), drop bias LR by 1/10 at epoch 45
weight_sched = Schedule([22, 44, 65, 90, 97], (1 / 250.)**(1 / 3.))
opt_gdm = GradientDescentMomentum(0.01,
0.9,
wdecay=0.005,
schedule=weight_sched)
opt_biases = GradientDescentMomentum(0.04,
1.0,
schedule=Schedule([130], .1))
opt = MultiOptimizer({'default': opt_gdm, 'Bias': opt_biases})
# configure callbacks
valmetric = TopKMisclassification(k=5)
callbacks = Callbacks(model,
train,
eval_set=val,
metric=valmetric,
**args.callback_args)
cost = GeneralizedCost(costfunc=CrossEntropyMulti())
#flag = input("Press Enter if you want to begin training process.")
print "Training network..."
model.fit(train,
optimizer=opt,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
mets = model.eval(test, metric=valmetric)
print 'Validation set metrics:'
print 'LogLoss: %.2f, Accuracy: %.1f %%0 (Top-1), %.1f %% (Top-5)' % (
mets[0], (1.0 - mets[1]) * 100, (1.0 - mets[2]) * 100)
test.exit_batch_provider()
val.exit_batch_provider()
train.exit_batch_provider()
def test_model_serialize(backend_default, data):
(X_train, y_train), (X_test, y_test), nclass = load_mnist(path=data)
train_set = ArrayIterator([X_train, X_train],
y_train,
nclass=nclass,
lshape=(1, 28, 28))
init_norm = Gaussian(loc=0.0, scale=0.01)
# initialize model
path1 = Sequential([
Conv((5, 5, 16),
init=init_norm,
bias=Constant(0),
activation=Rectlin()),
Pooling(2),
Affine(nout=20, init=init_norm, bias=init_norm, activation=Rectlin())
])
path2 = Sequential([
Affine(nout=100,
init=init_norm,
bias=Constant(0),
activation=Rectlin()),
Dropout(keep=0.5),
Affine(nout=20, init=init_norm, bias=init_norm, activation=Rectlin())
])
layers = [
MergeMultistream(layers=[path1, path2], merge="stack"),
Affine(nout=20, init=init_norm, batch_norm=True, activation=Rectlin()),
Affine(nout=10, init=init_norm, activation=Logistic(shortcut=True))
]
tmp_save = 'test_model_serialize_tmp_save.pickle'
mlp = Model(layers=layers)
mlp.optimizer = GradientDescentMomentum(learning_rate=0.1,
momentum_coef=0.9)
mlp.cost = GeneralizedCost(costfunc=CrossEntropyBinary())
mlp.initialize(train_set, cost=mlp.cost)
n_test = 3
num_epochs = 3
# Train model for num_epochs and n_test batches
for epoch in range(num_epochs):
for i, (x, t) in enumerate(train_set):
x = mlp.fprop(x)
delta = mlp.cost.get_errors(x, t)
mlp.bprop(delta)
mlp.optimizer.optimize(mlp.layers_to_optimize, epoch=epoch)
if i > n_test:
break
# Get expected outputs of n_test batches and states of all layers
outputs_exp = []
pdicts_exp = [l.get_params_serialize() for l in mlp.layers_to_optimize]
for i, (x, t) in enumerate(train_set):
outputs_exp.append(mlp.fprop(x, inference=True))
if i > n_test:
break
# Serialize model
mlp.save_params(tmp_save, keep_states=True)
# Load model
mlp = Model(tmp_save)
mlp.initialize(train_set)
outputs = []
pdicts = [l.get_params_serialize() for l in mlp.layers_to_optimize]
for i, (x, t) in enumerate(train_set):
outputs.append(mlp.fprop(x, inference=True))
if i > n_test:
break
# Check outputs, states, and params are the same
for output, output_exp in zip(outputs, outputs_exp):
assert np.allclose(output.get(), output_exp.get())
for pd, pd_exp in zip(pdicts, pdicts_exp):
for s, s_e in zip(pd['states'], pd_exp['states']):
if isinstance(s, list): # this is the batch norm case
for _s, _s_e in zip(s, s_e):
assert np.allclose(_s, _s_e)
else:
assert np.allclose(s, s_e)
for p, p_e in zip(pd['params'], pd_exp['params']):
assert type(p) == type(p_e)
if isinstance(p, list): # this is the batch norm case
for _p, _p_e in zip(p, p_e):
assert np.allclose(_p, _p_e)
elif isinstance(p, np.ndarray):
assert np.allclose(p, p_e)
else:
assert p == p_e
os.remove(tmp_save)
print D_layers
print G_layers
layers = myGenerativeAdversarial(generator=G_layers, discriminator=D_layers)
#discriminator=Sequential(D_layers, name="Discriminator"))
print 'layers defined'
print layers
# setup optimizer
# optimizer = RMSProp(learning_rate=1e-4, decay_rate=0.9, epsilon=1e-8)
optimizer = GradientDescentMomentum(learning_rate=1e-3, momentum_coef = 0.9)
#optimizer = Adam(learning_rate=1e-3)
# setup cost function as Binary CrossEntropy
#cost = GeneralizedGANCost(costfunc=GANCost(func="wasserstein"))
#cost = GeneralizedGANCost(costfunc=Multicost[GANCost(func="wasserstein"), MeanSquared, MeanSquared])
cost = Multicost(costs=[GeneralizedGANCost(costfunc=GANCost(func="wasserstein")), GeneralizedCost(costfunc=MeanSquared), GeneralizedCost(costfunc=MeanSquared)])
nb_epochs = 15
latent_size = 200
# initialize model
noise_dim = (latent_size)
gan = myGAN(layers=layers, noise_dim=noise_dim, dataset=train_set, k=5, wgan_param_clamp=0.9)
# configure callbacks
callbacks = Callbacks(gan, eval_set=valid_set)
callbacks.add_callback(GANCostCallback())
#callbacks.add_save_best_state_callback("./best_state.pkl")
print 'starting training'
# run fit
gan.fit(train_set, num_epochs=nb_epochs, optimizer=optimizer,
cost=cost, callbacks=callbacks)
# setting model layers for AE1
encoder1 = Affine(nout=config.encoder_size[0], init=init_norm,
activation=Logistic(), name='encoder1')
decoder1 = Affine(nout=image_size, init=init_norm, activation=Logistic(),
name='decoder1')
encoder2 = Affine(nout=config.encoder_size[1], init=init_norm,
activation=Logistic(), name='encoder2')
decoder2 = Affine(nout=config.encoder_size[0], init=init_norm,
activation=Logistic(), name='decoder2')
encoder3 = Affine(nout=config.encoder_size[2], init=init_norm,
activation=Logistic(), name='encoder3')
decoder3 = Affine(nout=config.encoder_size[1], init=init_norm,
activation=Logistic(), name='decoder3')
classifier = Affine(nout=config.ydim, init=init_norm, activation=Softmax())
cost_reconst = GeneralizedCost(costfunc=SumSquared())
cost_classification = GeneralizedCost(costfunc=CrossEntropyMulti())
# Setting model layers for AE1
AE1 = Model([encoder1, decoder1])
AE1.cost = cost_reconst
AE1.initialize(data, cost_reconst)
# AE1.optimizer = optimizer_default
measure_time(data, AE1, config, 'AE1')
# Setting model layers for AE2
# It has an extra encoder layer compared to what AE should really be. This is
# done to avoid saving the outputs for each AE.
AE2_mimic = Model([encoder1, encoder2, decoder2])
AE2_mimic.cost = cost_reconst
AE2_mimic.initialize(data, cost_reconst)
def benchmark(self):
for d in self.devices:
b = d if (self.backends is None) or (
"mkl" not in self.backends) else "mkl"
print("Use {} as backend.".format(b))
# Common suffix
suffix = "neon_{}_{}_{}by{}_{}".format(b, self.dataset,
self.resize_size[0],
self.resize_size[1],
self.preprocessing)
# Set up backend
# backend: 'cpu' for single cpu, 'mkl' for cpu using mkl library, and 'gpu' for gpu
be = gen_backend(backend=b,
batch_size=self.batch_size,
rng_seed=542,
datatype=np.float32)
# Prepare training/validation/testing sets
neon_train_set = ArrayIterator(X=np.asarray(
[t.flatten().astype('float32') / 255 for t in self.x_train]),
y=np.asarray(self.y_train),
make_onehot=True,
nclass=self.class_num,
lshape=(3, self.resize_size[0],
self.resize_size[1]))
neon_valid_set = ArrayIterator(X=np.asarray(
[t.flatten().astype('float32') / 255 for t in self.x_valid]),
y=np.asarray(self.y_valid),
make_onehot=True,
nclass=self.class_num,
lshape=(3, self.resize_size[0],
self.resize_size[1]))
neon_test_set = ArrayIterator(X=np.asarray([
t.flatten().astype('float32') / 255 for t in self.testImages
]),
y=np.asarray(self.testLabels),
make_onehot=True,
nclass=self.class_num,
lshape=(3, self.resize_size[0],
self.resize_size[1]))
# Initialize model object
self.neon_model = SelfModel(layers=self.constructCNN())
# Costs
neon_cost = GeneralizedCost(costfunc=CrossEntropyMulti())
# Model summary
self.neon_model.initialize(neon_train_set, neon_cost)
print(self.neon_model)
# Learning rules
neon_optimizer = SGD(0.01,
momentum_coef=0.9,
schedule=ExpSchedule(0.2))
# neon_optimizer = RMSProp(learning_rate=0.0001, decay_rate=0.95)
# # Benchmark for 20 minibatches
# d[b] = self.neon_model.benchmark(neon_train_set, cost=neon_cost, optimizer=neon_optimizer)
# Reset model
# self.neon_model = None
# self.neon_model = Model(layers=layers)
# self.neon_model.initialize(neon_train_set, neon_cost)
# Callbacks: validate on validation set
callbacks = Callbacks(
self.neon_model,
eval_set=neon_valid_set,
metric=Misclassification(3),
output_file="./saved_data/{}/{}/callback_data_{}.h5".format(
self.network_type, d, suffix))
callbacks.add_callback(
SelfCallback(eval_set=neon_valid_set,
test_set=neon_test_set,
epoch_freq=1))
# Fit
start = time.time()
self.neon_model.fit(neon_train_set,
optimizer=neon_optimizer,
num_epochs=self.epoch_num,
cost=neon_cost,
callbacks=callbacks)
print("Neon training finishes in {:.2f} seconds.".format(
time.time() - start))
# Result
# results = self.neon_model.get_outputs(neon_valid_set)
# Print error on validation set
start = time.time()
neon_error_mis = self.neon_model.eval(
neon_valid_set, metric=Misclassification()) * 100
print(
'Misclassification error = {:.1f}%. Finished in {:.2f} seconds.'
.format(neon_error_mis[0],
time.time() - start))
# start = time.time()
# neon_error_top3 = self.neon_model.eval(neon_valid_set, metric=TopKMisclassification(3))*100
# print('Top 3 Misclassification error = {:.1f}%. Finished in {:.2f} seconds.'.format(neon_error_top3[2], time.time() - start))
# start = time.time()
# neon_error_top5 = self.neon_model.eval(neon_valid_set, metric=TopKMisclassification(5))*100
# print('Top 5 Misclassification error = {:.1f}%. Finished in {:.2f} seconds.'.format(neon_error_top5[2], time.time() - start))
self.neon_model.save_params("./saved_models/{}/{}/{}.prm".format(
self.network_type, d, suffix))
# Print error on test set
start = time.time()
neon_error_mis_t = self.neon_model.eval(
neon_test_set, metric=Misclassification()) * 100
print(
'Misclassification error = {:.1f}% on test set. Finished in {:.2f} seconds.'
.format(neon_error_mis_t[0],
time.time() - start))
# start = time.time()
# neon_error_top3_t = self.neon_model.eval(neon_test_set, metric=TopKMisclassification(3))*100
# print('Top 3 Misclassification error = {:.1f}% on test set. Finished in {:.2f} seconds.'.format(neon_error_top3_t[2], time.time() - start))
# start = time.time()
# neon_error_top5_t = self.neon_model.eval(neon_test_set, metric=TopKMisclassification(5))*100
# print('Top 5 Misclassification error = {:.1f}% on test set. Finished in {:.2f} seconds.'.format(neon_error_top5_t[2], time.time() - start))
cleanup_backend()
self.neon_model = None
def main():
# Get command-line parameters
parser = get_p1b3_parser()
args = parser.parse_args()
#print('Args:', args)
# Get parameters from configuration file
fileParameters = p1b3.read_config_file(args.config_file)
#print ('Params:', fileParameters)
# Correct for arguments set by default by neon parser
# (i.e. instead of taking the neon parser default value fall back to the config file,
# if effectively the command-line was used, then use the command-line value)
# This applies to conflictive parameters: batch_size, epochs and rng_seed
if not any("--batch_size" in ag or "-z" in ag for ag in sys.argv):
args.batch_size = fileParameters['batch_size']
if not any("--epochs" in ag or "-e" in ag for ag in sys.argv):
args.epochs = fileParameters['epochs']
if not any("--rng_seed" in ag or "-r" in ag for ag in sys.argv):
args.rng_seed = fileParameters['rng_seed']
# Consolidate parameter set. Command-line parameters overwrite file configuration
gParameters = p1_common.args_overwrite_config(args, fileParameters)
print('Params:', gParameters)
# Determine verbosity level
loggingLevel = logging.DEBUG if args.verbose else logging.INFO
logging.basicConfig(level=loggingLevel, format='')
# Construct extension to save model
ext = p1b3.extension_from_parameters(gParameters, '.neon')
# Get default parameters for initialization and optimizer functions
kerasDefaults = p1_common.keras_default_config()
seed = gParameters['rng_seed']
# Build dataset loader object
loader = p1b3.DataLoader(
seed=seed,
dtype=gParameters['datatype'],
val_split=gParameters['validation_split'],
test_cell_split=gParameters['test_cell_split'],
cell_features=gParameters['cell_features'],
drug_features=gParameters['drug_features'],
feature_subsample=gParameters['feature_subsample'],
scaling=gParameters['scaling'],
scramble=gParameters['scramble'],
min_logconc=gParameters['min_logconc'],
max_logconc=gParameters['max_logconc'],
subsample=gParameters['subsample'],
category_cutoffs=gParameters['category_cutoffs'])
# Re-generate the backend after consolidating parsing and file config
gen_backend(backend=args.backend,
rng_seed=seed,
device_id=args.device_id,
batch_size=gParameters['batch_size'],
datatype=gParameters['datatype'],
max_devices=args.max_devices,
compat_mode=args.compat_mode)
# Initialize weights and learning rule
initializer_weights = p1_common_neon.build_initializer(
gParameters['initialization'], kerasDefaults, seed)
initializer_bias = p1_common_neon.build_initializer(
'constant', kerasDefaults, 0.)
activation = p1_common_neon.get_function(gParameters['activation'])()
# Define model architecture
layers = []
reshape = None
if 'dense' in gParameters: # Build dense layers
for layer in gParameters['dense']:
if layer:
layers.append(
Affine(nout=layer,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
if gParameters['drop']:
layers.append(Dropout(keep=(1 - gParameters['drop'])))
else: # Build convolutional layers
reshape = (1, loader.input_dim, 1)
layer_list = list(range(0, len(gParameters['conv']), 3))
for l, i in enumerate(layer_list):
nb_filter = gParameters['conv'][i]
filter_len = gParameters['conv'][i + 1]
stride = gParameters['conv'][i + 2]
# print(nb_filter, filter_len, stride)
# fshape: (height, width, num_filters).
layers.append(
Conv((1, filter_len, nb_filter),
strides={
'str_h': 1,
'str_w': stride
},
init=initializer_weights,
activation=activation))
if gParameters['pool']:
layers.append(Pooling((1, gParameters['pool'])))
layers.append(
Affine(nout=1,
init=initializer_weights,
bias=initializer_bias,
activation=neon.transforms.Identity()))
# Build model
model = Model(layers=layers)
# Define neon data iterators
train_samples = int(loader.n_train)
val_samples = int(loader.n_val)
if 'train_samples' in gParameters:
train_samples = gParameters['train_samples']
if 'val_samples' in gParameters:
val_samples = gParameters['val_samples']
train_iter = ConcatDataIter(loader,
ndata=train_samples,
lshape=reshape,
datatype=gParameters['datatype'])
val_iter = ConcatDataIter(loader,
partition='val',
ndata=val_samples,
lshape=reshape,
datatype=gParameters['datatype'])
# Define cost and optimizer
cost = GeneralizedCost(p1_common_neon.get_function(gParameters['loss'])())
optimizer = p1_common_neon.build_optimizer(gParameters['optimizer'],
gParameters['learning_rate'],
kerasDefaults)
callbacks = Callbacks(model, eval_set=val_iter,
eval_freq=1) #**args.callback_args)
model.fit(train_iter,
optimizer=optimizer,
num_epochs=gParameters['epochs'],
cost=cost,
callbacks=callbacks)
from neon.backends import gen_backend
import bot_params as params
import replay_memory as mem
from enemydetector1 import model, predict
params.batch_size = 64
be = gen_backend(backend='cpu', batch_size=params.batch_size)
dataset = mem.load()
opt_gdm = GradientDescentMomentum(learning_rate=0.01,
momentum_coef=0.9,
stochastic_round=0)
cost = GeneralizedCost(costfunc=CrossEntropyMulti(scale=10))
(X_train, y_train), (X_test, y_test) = dataset.get_dataset()
print X_train.shape, y_train.shape, X_test.shape, y_test.shape
train_set = ArrayIterator(X=X_train, y=y_train, nclass=dataset.nclass, lshape=dataset.shape, make_onehot=False)
test = ArrayIterator(X=X_test, y=y_test, nclass=dataset.nclass, lshape=dataset.shape, make_onehot=False)
callbacks = Callbacks(model, eval_set=test, eval_freq=1,)
model.fit(train_set, optimizer=opt_gdm, num_epochs=2, cost=cost, callbacks=callbacks)
model.save_params(params.weigths_path)
def test_example(i):
val = predict(X_train[i])
def create_network_lrn():
init1 = Gaussian(scale=0.01)
init2 = Gaussian(scale=0.005)
layers = [
Conv((11, 11, 96),
padding=0,
strides=4,
init=init1,
bias=Constant(0),
activation=Rectlin(),
name='conv1'),
Pooling(3, strides=2, name='pool1'),
LRN(5, ascale=0.0001, bpower=0.75, name='norm1'),
Conv((5, 5, 256),
padding=2,
init=init1,
bias=Constant(1.0),
activation=Rectlin(),
name='conv2'),
Pooling(3, strides=2, name='pool2'),
LRN(5, ascale=0.0001, bpower=0.75, name='norm2'),
Conv((3, 3, 384),
padding=1,
init=init1,
bias=Constant(0),
activation=Rectlin(),
name='conv3'),
Conv((3, 3, 384),
padding=1,
init=init1,
bias=Constant(1.0),
activation=Rectlin(),
name='conv4'),
Conv((3, 3, 256),
padding=1,
init=init1,
bias=Constant(1.0),
activation=Rectlin(),
name='conv5'),
Pooling(3, strides=2, name='pool5'),
Affine(nout=4096,
init=init2,
bias=Constant(1.0),
activation=Rectlin(),
name='fc6'),
Dropout(keep=0.5, name='drop6'),
Affine(nout=4096,
init=init2,
bias=Constant(1.0),
activation=Rectlin(),
name='fc7'),
Dropout(keep=0.5, name='drop7'),
Affine(nout=1000,
init=init1,
bias=Constant(0.0),
activation=Softmax(),
name='fc8')
]
return Model(layers=layers), GeneralizedCost(costfunc=CrossEntropyMulti())
def main():
# larger batch sizes may not fit on GPU
parser = NeonArgparser(__doc__, default_overrides={'batch_size': 4})
parser.add_argument("--bench", action="store_true", help="run benchmark instead of training")
parser.add_argument("--num_classes", type=int, default=12, help="number of classes in the annotation")
parser.add_argument("--height", type=int, default=256, help="image height")
parser.add_argument("--width", type=int, default=512, help="image width")
args = parser.parse_args(gen_be=False)
# check that image dimensions are powers of 2
if((args.height & (args.height - 1)) != 0):
raise TypeError("Height must be a power of 2.")
if((args.width & (args.width - 1)) != 0):
raise TypeError("Width must be a power of 2.")
(c, h, w) = (args.num_classes, args.height, args.width)
# need to use the backend with the new upsampling layer implementation
be = NervanaGPU_Upsample(rng_seed=args.rng_seed,
device_id=args.device_id)
# set batch size
be.bsz = args.batch_size
# couple backend to global neon object
NervanaObject.be = be
shape = dict(channel_count=3, height=h, width=w, subtract_mean=False)
train_params = ImageParams(center=True, flip=False,
scale_min=min(h, w), scale_max=min(h, w),
aspect_ratio=0, **shape)
test_params = ImageParams(center=True, flip=False,
scale_min=min(h, w), scale_max=min(h, w),
aspect_ratio=0, **shape)
common = dict(target_size=h*w, target_conversion='read_contents',
onehot=False, target_dtype=np.uint8, nclasses=args.num_classes)
train_set = PixelWiseImageLoader(set_name='train', repo_dir=args.data_dir,
media_params=train_params,
shuffle=False, subset_percent=100,
index_file=os.path.join(args.data_dir, 'train_images.csv'),
**common)
val_set = PixelWiseImageLoader(set_name='val', repo_dir=args.data_dir,media_params=test_params,
index_file=os.path.join(args.data_dir, 'val_images.csv'), **common)
# initialize model object
layers = gen_model(c, h, w)
segnet_model = Model(layers=layers)
# configure callbacks
callbacks = Callbacks(segnet_model, eval_set=val_set, **args.callback_args)
opt_gdm = GradientDescentMomentum(1.0e-6, 0.9, wdecay=0.0005, schedule=Schedule())
opt_biases = GradientDescentMomentum(2.0e-6, 0.9, schedule=Schedule())
opt_bn = GradientDescentMomentum(1.0e-6, 0.9, schedule=Schedule())
opt = MultiOptimizer({'default': opt_gdm, 'Bias': opt_biases, 'BatchNorm': opt_bn})
cost = GeneralizedCost(costfunc=CrossEntropyMulti())
if args.bench:
segnet_model.initialize(train_set, cost=cost)
segnet_model.benchmark(train_set, cost=cost, optimizer=opt)
sys.exit(0)
else:
segnet_model.fit(train_set, optimizer=opt, num_epochs=args.epochs, cost=cost, callbacks=callbacks)
# get the trained segnet model outputs for valisation set
outs_val = segnet_model.get_outputs(val_set)
with open('outputs.pkl', 'w') as fid:
pickle.dump(outs_val, fid, -1)
wdecay=args.weight_decay,
schedule=weight_sched,
stochastic_round=args.rounding)
opt_biases = GradientDescentMomentum(args.rate_init[1],
args.momentum[1],
schedule=weight_sched,
stochastic_round=args.rounding)
opt_fixed = GradientDescentMomentum(0.0, 1.0, wdecay=0.0)
opt = MultiOptimizer({
'default': opt_gdm,
'Bias': opt_biases,
'DOG': opt_fixed
})
# configure cost and test metrics
cost = GeneralizedCost(costfunc=(CrossEntropyBinary() \
if train.parser.independent_labels else CrossEntropyMulti()))
metric = EMMetric(
oshape=test.parser.oshape,
use_softmax=not train.parser.independent_labels) if test else None
# configure callbacks
if not args.neon_progress:
args.callback_args['progress_bar'] = False
callbacks = Callbacks(model,
eval_set=test,
metric=metric,
**args.callback_args)
if not args.neon_progress:
callbacks.add_callback(EMEpochCallback(
args.callback_args['eval_freq'], train.nmacrobatches),
insert_pos=None)
train_set = ArrayIterator(X=X_train, y=y_train, make_onehot=False)
val_set = ArrayIterator(X=X_val, y=y_val, make_onehot=False)
# setup weight initialization function
init = Uniform(-1, 1)
# setup layers
layers = [
BinaryAffine(nout=4096, init=init, batch_norm=True, activation=Sign()),
BinaryAffine(nout=4096, init=init, batch_norm=True, activation=Sign()),
BinaryAffine(nout=4096, init=init, batch_norm=True, activation=Sign()),
BinaryAffine(nout=2, init=init, batch_norm=True, activation=Identity())
]
# setup cost function as Square Hinge Loss
cost = GeneralizedCost(costfunc=SquareHingeLoss())
# setup optimizer
LR_start = 1.65e-2
def ShiftAdaMax_with_Scale(LR=1):
return ShiftAdaMax(learning_rate=LR_start * LR,
schedule=ShiftSchedule(2, shift_size=1))
optimizer = MultiOptimizer({
'default': ShiftAdaMax_with_Scale(),
'BinaryLinear_0': ShiftAdaMax_with_Scale(57.038),
'BinaryLinear_1': ShiftAdaMax_with_Scale(73.9008),
'BinaryLinear_2': ShiftAdaMax_with_Scale(73.9008),
def main():
# Get command-line parameters
parser = get_p1b2_parser()
args = parser.parse_args()
#print('Args:', args)
# Get parameters from configuration file
fileParameters = p1b2.read_config_file(args.config_file)
#print ('Params:', fileParameters)
# Correct for arguments set by default by neon parser
# (i.e. instead of taking the neon parser default value fall back to the config file,
# if effectively the command-line was used, then use the command-line value)
# This applies to conflictive parameters: batch_size, epochs and rng_seed
if not any("--batch_size" in ag or "-z" in ag for ag in sys.argv):
args.batch_size = fileParameters['batch_size']
if not any("--epochs" in ag or "-e" in ag for ag in sys.argv):
args.epochs = fileParameters['epochs']
if not any("--rng_seed" in ag or "-r" in ag for ag in sys.argv):
args.rng_seed = fileParameters['rng_seed']
# Consolidate parameter set. Command-line parameters overwrite file configuration
gParameters = p1_common.args_overwrite_config(args, fileParameters)
print('Params:', gParameters)
# Determine verbosity level
loggingLevel = logging.DEBUG if args.verbose else logging.INFO
logging.basicConfig(level=loggingLevel, format='')
# Construct extension to save model
ext = p1b2.extension_from_parameters(gParameters, '.neon')
# Get default parameters for initialization and optimizer functions
kerasDefaults = p1_common.keras_default_config()
seed = gParameters['rng_seed']
# Load dataset
#(X_train, y_train), (X_test, y_test) = p1b2.load_data(gParameters, seed)
(X_train, y_train), (X_val,
y_val), (X_test,
y_test) = p1b2.load_data(gParameters, seed)
print("Shape X_train: ", X_train.shape)
print("Shape X_val: ", X_val.shape)
print("Shape X_test: ", X_test.shape)
print("Shape y_train: ", y_train.shape)
print("Shape y_val: ", y_val.shape)
print("Shape y_test: ", y_test.shape)
print("Range X_train --> Min: ", np.min(X_train), ", max: ",
np.max(X_train))
print("Range X_val --> Min: ", np.min(X_val), ", max: ", np.max(X_val))
print("Range X_test --> Min: ", np.min(X_test), ", max: ", np.max(X_test))
print("Range y_train --> Min: ", np.min(y_train), ", max: ",
np.max(y_train))
print("Range y_val --> Min: ", np.min(y_val), ", max: ", np.max(y_val))
print("Range y_test --> Min: ", np.min(y_test), ", max: ", np.max(y_test))
input_dim = X_train.shape[1]
num_classes = int(np.max(y_train)) + 1
output_dim = num_classes # The backend will represent the classes using one-hot representation (but requires an integer class as input !)
# Re-generate the backend after consolidating parsing and file config
gen_backend(backend=args.backend,
rng_seed=seed,
device_id=args.device_id,
batch_size=gParameters['batch_size'],
datatype=gParameters['data_type'],
max_devices=args.max_devices,
compat_mode=args.compat_mode)
train = ArrayIterator(X=X_train, y=y_train, nclass=num_classes)
val = ArrayIterator(X=X_val, y=y_val, nclass=num_classes)
test = ArrayIterator(X=X_test, y=y_test, nclass=num_classes)
# Initialize weights and learning rule
initializer_weights = p1_common_neon.build_initializer(
gParameters['initialization'], kerasDefaults, seed)
initializer_bias = p1_common_neon.build_initializer(
'constant', kerasDefaults, 0.)
activation = p1_common_neon.get_function(gParameters['activation'])()
# Define MLP architecture
layers = []
reshape = None
for layer in gParameters['dense']:
if layer:
layers.append(
Affine(nout=layer,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
if gParameters['dropout']:
layers.append(Dropout(keep=(1 - gParameters['dropout'])))
layers.append(
Affine(nout=output_dim,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
# Build MLP model
mlp = Model(layers=layers)
# Define cost and optimizer
cost = GeneralizedCost(p1_common_neon.get_function(gParameters['loss'])())
optimizer = p1_common_neon.build_optimizer(gParameters['optimizer'],
gParameters['learning_rate'],
kerasDefaults)
callbacks = Callbacks(mlp, eval_set=val, metric=Accuracy(), eval_freq=1)
# Seed random generator for training
np.random.seed(seed)
mlp.fit(train,
optimizer=optimizer,
num_epochs=gParameters['epochs'],
cost=cost,
callbacks=callbacks)
# model save
#save_fname = "model_mlp_W_" + ext
#mlp.save_params(save_fname)
# Evalute model on test set
print('Model evaluation by neon: ', mlp.eval(test, metric=Accuracy()))
y_pred = mlp.get_outputs(test)
#print ("Shape y_pred: ", y_pred.shape)
scores = p1b2.evaluate_accuracy(p1_common.convert_to_class(y_pred), y_test)
print('Evaluation on test data:', scores)
momentum_coef=0.9,
stochastic_round=args.rounding)
elif args.datatype in [np.float16]:
opt_gdm = GradientDescentMomentum(learning_rate=0.01 / cost_scale,
momentum_coef=0.9,
stochastic_round=args.rounding)
bn = True
layers = [Conv((5, 5, 16), init=init_uni, activation=Rectlin(), batch_norm=bn),
Pooling((2, 2)),
Conv((5, 5, 32), init=init_uni, activation=Rectlin(), batch_norm=bn),
Pooling((2, 2)),
Affine(nout=500, init=init_uni, activation=Rectlin(), batch_norm=bn),
Affine(nout=10, init=init_uni, activation=Softmax())]
if args.datatype in [np.float32, np.float64]:
cost = GeneralizedCost(costfunc=CrossEntropyMulti())
elif args.datatype in [np.float16]:
cost = GeneralizedCost(costfunc=CrossEntropyMulti(scale=cost_scale))
model = Model(layers=layers)
# configure callbacks
callbacks = Callbacks(model, eval_set=test, **args.callback_args)
model.fit(train, optimizer=opt_gdm, num_epochs=num_epochs,
cost=cost, callbacks=callbacks)
error_rate = model.eval(test, metric=Misclassification())
neon_logger.display('Misclassification error = %.1f%%' % (error_rate * 100))
def main():
# Get command-line parameters
parser = get_p1b1_parser()
args = parser.parse_args()
#print('Args:', args)
# Get parameters from configuration file
fileParameters = p1b1.read_config_file(args.config_file)
#print ('Params:', fileParameters)
# Correct for arguments set by default by neon parser
# (i.e. instead of taking the neon parser default value fall back to the config file,
# if effectively the command-line was used, then use the command-line value)
# This applies to conflictive parameters: batch_size, epochs and rng_seed
if not any("--batch_size" in ag or "-z" in ag for ag in sys.argv):
args.batch_size = fileParameters['batch_size']
if not any("--epochs" in ag or "-e" in ag for ag in sys.argv):
args.epochs = fileParameters['epochs']
if not any("--rng_seed" in ag or "-r" in ag for ag in sys.argv):
args.rng_seed = fileParameters['rng_seed']
# Consolidate parameter set. Command-line parameters overwrite file configuration
gParameters = p1_common.args_overwrite_config(args, fileParameters)
print('Params:', gParameters)
# Determine verbosity level
loggingLevel = logging.DEBUG if args.verbose else logging.INFO
logging.basicConfig(level=loggingLevel, format='')
# Construct extension to save model
ext = p1b1.extension_from_parameters(gParameters, '.neon')
# Get default parameters for initialization and optimizer functions
kerasDefaults = p1_common.keras_default_config()
seed = gParameters['rng_seed']
# Load dataset
X_train, X_val, X_test = p1b1.load_data(gParameters, seed)
print("Shape X_train: ", X_train.shape)
print("Shape X_val: ", X_val.shape)
print("Shape X_test: ", X_test.shape)
print("Range X_train --> Min: ", np.min(X_train), ", max: ",
np.max(X_train))
print("Range X_val --> Min: ", np.min(X_val), ", max: ", np.max(X_val))
print("Range X_test --> Min: ", np.min(X_test), ", max: ", np.max(X_test))
input_dim = X_train.shape[1]
output_dim = input_dim
# Re-generate the backend after consolidating parsing and file config
gen_backend(backend=args.backend,
rng_seed=seed,
device_id=args.device_id,
batch_size=gParameters['batch_size'],
datatype=gParameters['datatype'],
max_devices=args.max_devices,
compat_mode=args.compat_mode)
# Set input and target to X_train
train = ArrayIterator(X_train)
val = ArrayIterator(X_val)
test = ArrayIterator(X_test)
# Initialize weights and learning rule
initializer_weights = p1_common_neon.build_initializer(
gParameters['initialization'], kerasDefaults)
initializer_bias = p1_common_neon.build_initializer(
'constant', kerasDefaults, 0.)
activation = p1_common_neon.get_function(gParameters['activation'])()
# Define Autoencoder architecture
layers = []
reshape = None
# Autoencoder
layers_params = gParameters['dense']
if layers_params != None:
if type(layers_params) != list:
layers_params = list(layers_params)
# Encoder Part
for i, l in enumerate(layers_params):
layers.append(
Affine(nout=l,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
# Decoder Part
for i, l in reversed(list(enumerate(layers_params))):
if i < len(layers) - 1:
layers.append(
Affine(nout=l,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
layers.append(
Affine(nout=output_dim,
init=initializer_weights,
bias=initializer_bias,
activation=activation))
# Build Autoencoder model
ae = Model(layers=layers)
# Define cost and optimizer
cost = GeneralizedCost(p1_common_neon.get_function(gParameters['loss'])())
optimizer = p1_common_neon.build_optimizer(gParameters['optimizer'],
gParameters['learning_rate'],
kerasDefaults)
callbacks = Callbacks(ae, eval_set=val, eval_freq=1)
# Seed random generator for training
np.random.seed(seed)
ae.fit(train,
optimizer=optimizer,
num_epochs=gParameters['epochs'],
cost=cost,
callbacks=callbacks)
# model save
#save_fname = "model_ae_W" + ext
#ae.save_params(save_fname)
# Compute errors
X_pred = ae.get_outputs(test)
scores = p1b1.evaluate_autoencoder(X_pred, X_test)
print('Evaluation on test data:', scores)
diff = X_pred - X_test
# Plot histogram of errors comparing input and output of autoencoder
plt.hist(diff.ravel(), bins='auto')
plt.title("Histogram of Errors with 'auto' bins")
plt.savefig('histogram_neon.png')
DeepBiRNN(hidden_size,
init=glorot,
activation=Rectlinclip(),
batch_norm=True,
reset_cells=True,
depth=depth),
Affine(hidden_size, init=glorot, activation=Rectlinclip()),
Affine(nout=nout, init=glorot, activation=Identity())
]
model = Model(layers=layers)
opt = GradientDescentMomentumNesterov(learning_rate,
momentum,
gradient_clip_norm=gradient_clip_norm,
stochastic_round=False)
callbacks = Callbacks(model, eval_set=dev, **args.callback_args)
# Print validation set word error rate at the end of every epoch
pcb = WordErrorRateCallback(dev, argmax_decoder, max_tscrpt_len, epoch_freq=1)
callbacks.add_callback(pcb)
cost = GeneralizedCost(costfunc=CTC(max_tscrpt_len, nout=nout))
# Fit the model
model.fit(train,
optimizer=opt,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
def main():
parser = NeonArgparser(__doc__)
args = parser.parse_args(gen_be=False)
#mat_data = sio.loadmat('../data/timeseries/02_timeseries.mat')
#ts = V1TimeSeries(mat_data['timeseries'], mat_data['stim'], binning=10)
seq_len = 30
hidden = 20
be = gen_backend(**extract_valid_args(args, gen_backend))
kohn = KohnV1Dataset(path='../tmp/')
kohn.gen_iterators(seq_len)
import pdb; pdb.set_trace()
train_spike_set = V1IteratorSequence(ts.train, seq_len, return_sequences=False)
valid_spike_set = V1IteratorSequence(ts.test, seq_len, return_sequences=False)
init = GlorotUniform()
# dataset = MNIST(path=args.data_dir)
# (X_train, y_train), (X_test, y_test), nclass = dataset.load_data()
# train_set = ArrayIterator([X_train, X_train], y_train, nclass=nclass, lshape=(1, 28, 28))
# valid_set = ArrayIterator([X_test, X_test], y_test, nclass=nclass, lshape=(1, 28, 28))
# # weight initialization
# init_norm = Gaussian(loc=0.0, scale=0.01)
# # initialize model
# path1 = Sequential(layers=[Affine(nout=100, init=init_norm, activation=Rectlin()),
# Affine(nout=100, init=init_norm, activation=Rectlin())])
# path2 = Sequential(layers=[Affine(nout=100, init=init_norm, activation=Rectlin()),
# Affine(nout=100, init=init_norm, activation=Rectlin())])
# layers = [MergeMultistream(layers=[path1, path2], merge="stack"),
# Affine(nout=10, init=init_norm, activation=Logistic(shortcut=True))]
spike_rnn_path = Sequential( layers = [
LSTM(hidden, init, activation=Logistic(),
gate_activation=Logistic(), reset_cells=False),
Dropout(keep=0.5),
LSTM(hidden, init, activation=Logistic(),
gate_activation=Logistic(), reset_cells=False),
#Dropout(keep=0.85),
RecurrentLast(),
Affine(train_set.nfeatures, init, bias=init, activation=Identity(), name='spike_in')])
stim_rnn_path = Sequential( layers = [
LSTM(hidden, init, activation=Logistic(),
gate_activation=Logistic(), reset_cells=False),
Dropout(keep=0.5),
RecurrentLast(),
Affine(1, init, bias=init, activation=Identity(), name='stim')])
layers = [
MergeMultiStream(
layers = [
spike_rnn_path,
stim_rnn_path],
merge="stack"),
Affine(train_set.nfeatures, init, bias=init, activation=Identity(), name='spike_out'),
Round()
]
model = Model(layers=layers)
sched = ExpSchedule(decay=0.7)
# cost = GeneralizedCost(SumSquared())
cost = GeneralizedCost(MeanSquared())
optimizer_two = RMSProp(stochastic_round=args.rounding)
optimizer_one = GradientDescentMomentum(learning_rate=0.1, momentum_coef=0.9, schedule=sched)
opt = MultiOptimizer({'default': optimizer_one,
'Bias': optimizer_two,
'special_linear': optimizer_two})
callbacks = Callbacks(model, eval_set=valid_set, **args.callback_args)
callbacks.add_hist_callback(filter_key = ['W'])
#callbacks.add_callback(MetricCallback(eval_set=valid_set, metric=FractionExplainedVariance(), epoch_freq=args.eval_freq))
#callbacks.add_callback(MetricCallback(eval_set=valid_set,metric=Accuracy(), epoch_freq=args.eval_freq))
model.fit(train_set,
optimizer=opt,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
train_output = model.get_outputs(
train_set).reshape(-1, train_set.nfeatures)
valid_output = model.get_outputs(
valid_set).reshape(-1, valid_set.nfeatures)
train_target = train_set.y_series
valid_target = valid_set.y_series
tfev = fev(train_output, train_target, train_set.mean)
vfev = fev(valid_output, valid_target, valid_set.mean)
neon_logger.display('Train FEV: %g, Valid FEV: %g' % (tfev, vfev))
# neon_logger.display('Train Mean: %g, Valid Mean: %g' % (train_set.mean, valid_set.mean))
plt.figure()
plt.plot(train_output[:, 0], train_output[
:, 1], 'bo', label='prediction')
plt.plot(train_target[:, 0], train_target[:, 1], 'r.', label='target')
plt.legend()
plt.title('Neon on training set')
plt.savefig('neon_series_training_output.png')
plt.figure()
plt.plot(valid_output[:, 0], valid_output[
:, 1], 'bo', label='prediction')
plt.plot(valid_target[:, 0], valid_target[:, 1], 'r.', label='target')
plt.legend()
plt.title('Neon on validation set')
plt.savefig('neon_series_validation_output.png')
Conv((3, 3, 192), name="D31", **convp1),
Conv((1, 1, 16), name="D32", **conv),
Conv((7, 7, 1),
name="D_out",
init=init,
batch_norm=False,
activation=Logistic(shortcut=False))
]
layers = GenerativeAdversarial(generator=Sequential(G_layers,
name="Generator"),
discriminator=Sequential(D_layers,
name="Discriminator"))
# setup cost function as CrossEntropy
cost = GeneralizedCost(costfunc=GANCost(func="modified"))
# setup optimizer
optimizer = Adam(learning_rate=0.0005, beta_1=0.5)
# initialize model
noise_dim = (2, 7, 7)
gan = GAN(layers=layers, noise_dim=noise_dim, k=args.kbatch)
# configure callbacks
callbacks = Callbacks(gan, eval_set=valid_set, **args.callback_args)
callbacks.add_callback(GANPlotCallback(filename=splitext(__file__)[0], hw=27))
# run fit
gan.fit(train_set,
optimizer=optimizer,
num_epochs=args.epochs,
def __init__(self, num_actions, args):
# remember parameters
self.num_actions = num_actions
self.batch_size = args.batch_size
self.discount_rate = args.discount_rate
self.history_length = args.history_length
self.screen_dim = (args.screen_height, args.screen_width)
self.clip_error = args.clip_error
self.min_reward = args.min_reward
self.max_reward = args.max_reward
self.batch_norm = args.batch_norm
# create Neon backend
self.be = gen_backend(backend=args.backend,
batch_size=args.batch_size,
rng_seed=args.random_seed,
device_id=args.device_id,
datatype=np.dtype(args.datatype).type,
stochastic_round=args.stochastic_round)
# prepare tensors once and reuse them
self.input_shape = (self.history_length, ) + self.screen_dim + (
self.batch_size, )
self.input = self.be.empty(self.input_shape)
self.input.lshape = self.input_shape # HACK: needed for convolutional networks
self.targets = self.be.empty((self.num_actions, self.batch_size))
# create model
layers = self._createLayers(num_actions)
self.model = Model(layers=layers)
self.cost = GeneralizedCost(costfunc=SumSquared())
# Bug fix
for l in self.model.layers.layers:
l.parallelism = 'Disabled'
self.model.initialize(self.input_shape[:-1], self.cost)
if args.optimizer == 'rmsprop':
self.optimizer = RMSProp(learning_rate=args.learning_rate,
decay_rate=args.decay_rate,
stochastic_round=args.stochastic_round)
elif args.optimizer == 'adam':
self.optimizer = Adam(learning_rate=args.learning_rate,
stochastic_round=args.stochastic_round)
elif args.optimizer == 'adadelta':
self.optimizer = Adadelta(decay=args.decay_rate,
stochastic_round=args.stochastic_round)
else:
assert false, "Unknown optimizer"
# create target model
self.target_steps = args.target_steps
self.train_iterations = 0
if self.target_steps:
self.target_model = Model(layers=self._createLayers(num_actions))
# Bug fix
for l in self.target_model.layers.layers:
l.parallelism = 'Disabled'
self.target_model.initialize(self.input_shape[:-1])
self.save_weights_prefix = args.save_weights_prefix
else:
self.target_model = self.model
self.callback = None
def main():
# setup the model and run for num_epochs saving the last state only
# this is at the top so that the be is generated
mlp = gen_model(args.backend)
# setup data iterators
(X_train, y_train), (X_test,
y_test), nclass = load_mnist(path=args.data_dir)
if args.backend == 'nervanacpu' or args.backend == 'cpu':
# limit data since cpu backend runs slower
train = DataIterator(X_train[:1000],
y_train[:1000],
nclass=nclass,
lshape=(1, 28, 28))
valid = DataIterator(X_test[:1000],
y_test[:1000],
nclass=nclass,
lshape=(1, 28, 28))
else:
train = DataIterator(X_train,
y_train,
nclass=nclass,
lshape=(1, 28, 28))
valid = DataIterator(X_test, y_test, nclass=nclass, lshape=(1, 28, 28))
# serialization related
cost = GeneralizedCost(costfunc=CrossEntropyBinary())
opt_gdm = GradientDescentMomentum(learning_rate=0.1, momentum_coef=0.9)
checkpoint_model_path = os.path.join('./', 'test_oneshot.pkl')
checkpoint_schedule = 1 # save at every step
callbacks = Callbacks(mlp, train)
callbacks.add_serialize_callback(checkpoint_schedule,
checkpoint_model_path,
history=2)
# run the fit all the way through saving a checkpoint e
mlp.fit(train,
optimizer=opt_gdm,
num_epochs=num_epochs,
cost=cost,
callbacks=callbacks)
# setup model with same random seed run epoch by epoch
# serializing and deserializing at each step
mlp = gen_model(args.backend)
cost = GeneralizedCost(costfunc=CrossEntropyBinary())
opt_gdm = GradientDescentMomentum(learning_rate=0.1, momentum_coef=0.9)
# reset data iterators
train.reset()
valid.reset()
checkpoint_model_path = os.path.join('./', 'test_manyshot.pkl')
checkpoint_schedule = 1 # save at evey step
callbacks = Callbacks(mlp, train)
callbacks.add_serialize_callback(checkpoint_schedule,
checkpoint_model_path,
history=num_epochs)
for epoch in range(num_epochs):
# _0 points to state at end of epoch 0
mlp.fit(train,
optimizer=opt_gdm,
num_epochs=epoch + 1,
cost=cost,
callbacks=callbacks)
# load saved file
prts = os.path.splitext(checkpoint_model_path)
fn = prts[0] + '_%d' % epoch + prts[1]
mlp.load_weights(fn) # load the saved weights
# compare test_oneshot_<num_epochs>.pkl to test_manyshot_<num_epochs>.pkl
try:
compare_model_pickles('test_oneshot_%d.pkl' % (num_epochs - 1),
'test_manyshot_%d.pkl' % (num_epochs - 1))
except:
print 'test failed....'
sys.exit(1)
# setup optimizer
opt_w = GradientDescentMomentum(0.001 * learning_rate_scale, 0.9, wdecay=0.0005)
opt_b = GradientDescentMomentum(0.002 * learning_rate_scale, 0.9)
optimizer = MultiOptimizer({'default': opt_w, 'Bias': opt_b})
# setup model
model = Model(layers=Tree([frcn_layers, bb_layers]))
# if training a new model, seed the Alexnet conv layers with pre-trained weights
# otherwise, just load the model file
if args.model_file is None:
load_imagenet_weights(model, args.data_dir)
cost = Multicost(costs=[GeneralizedCost(costfunc=CrossEntropyMulti()),
GeneralizedCostMask(costfunc=SmoothL1Loss())],
weights=[1, 1])
callbacks = Callbacks(model, **args.callback_args)
model.fit(train_set, optimizer=optimizer,
num_epochs=num_epochs, cost=cost, callbacks=callbacks)
print 'running eval on the training set...'
metric_train = model.eval(train_set, metric=ObjectDetection())
print 'Train: label accuracy - {}%, object deteciton SmoothL1Loss - {}'.format(
metric_train[0]*100,
metric_train[1])
import numpy as np
import replay_memory
import enemydetector1
try:
offset_memory = replay_memory.load(bot_params.offset_data_path)
print "offsets loaded"
except IOError:
offset_memory = replay_memory.OffsetMemory()
init_norm = Gaussian(loc=-0.1, scale=0.1)
layers = [Linear(1, init=init_norm), Bias(init=init_norm)]
mlp = Model(layers=layers)
cost = GeneralizedCost(costfunc=SumSquared())
optimizer = GradientDescentMomentum(0.5, momentum_coef=0.9)
try:
mlp.load_params(bot_params.aim_weights_path)
except IOError:
print "can't load aiming weights"
def get_offset_manual(predictions):
#enemy_pos = replay_memory.clean_values_toone(predictions)[0, 0]
x = 0.
c = 0
for i in range(0, len(predictions)):
if predictions[i] > 0.97:
x += (i + 1)
c += 1
def test_conv_rnn(backend_default):
train_shape = (1, 17, 142)
be = backend_default
inp = be.array(be.rng.randn(np.prod(train_shape), be.bsz))
delta = be.array(be.rng.randn(10, be.bsz))
init_norm = Gaussian(loc=0.0, scale=0.01)
bilstm = DeepBiLSTM(128,
init_norm,
activation=Rectlin(),
gate_activation=Rectlin(),
depth=1,
reset_cells=True)
birnn_1 = DeepBiRNN(128,
init_norm,
activation=Rectlin(),
depth=1,
reset_cells=True,
batch_norm=False)
birnn_2 = DeepBiRNN(128,
init_norm,
activation=Rectlin(),
depth=2,
reset_cells=True,
batch_norm=False)
bibnrnn = DeepBiRNN(128,
init_norm,
activation=Rectlin(),
depth=1,
reset_cells=True,
batch_norm=True)
birnnsum = DeepBiRNN(128,
init_norm,
activation=Rectlin(),
depth=1,
reset_cells=True,
batch_norm=False,
bi_sum=True)
rnn = Recurrent(128,
init=init_norm,
activation=Rectlin(),
reset_cells=True)
lstm = LSTM(128,
init_norm,
activation=Rectlin(),
gate_activation=Rectlin(),
reset_cells=True)
gru = GRU(128,
init_norm,
activation=Rectlin(),
gate_activation=Rectlin(),
reset_cells=True)
rlayers = [bilstm, birnn_1, birnn_2, bibnrnn, birnnsum, rnn, lstm, gru]
for rl in rlayers:
layers = [
Conv((2, 2, 4),
init=init_norm,
activation=Rectlin(),
strides=dict(str_h=2, str_w=4)),
Pooling(2, strides=2),
Conv((3, 3, 4),
init=init_norm,
batch_norm=True,
activation=Rectlin(),
strides=dict(str_h=1, str_w=2)),
rl,
RecurrentMean(),
Affine(nout=10, init=init_norm, activation=Rectlin()),
]
model = Model(layers=layers)
cost = GeneralizedCost(costfunc=CrossEntropyBinary())
model.initialize(train_shape, cost)
model.fprop(inp)
model.bprop(delta)
normalize=True,
contrast_normalize=False,
whiten=False)
train = dataset.train_iter
test = dataset.valid_iter
init_uni = Uniform(low=-0.1, high=0.1)
opt_gdm = GradientDescentMomentum(learning_rate=0.01, momentum_coef=0.9)
# set up the model layers
layers = [
Affine(nout=200, init=init_uni, activation=Rectlin()),
Affine(nout=10, init=init_uni, activation=Logistic(shortcut=True))
]
cost = GeneralizedCost(costfunc=CrossEntropyBinary())
mlp = Model(layers=layers)
# configure callbacks
callbacks = Callbacks(mlp, eval_set=test, **args.callback_args)
mlp.fit(train,
optimizer=opt_gdm,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
neon_logger.display('Misclassification error = %.1f%%' %
(mlp.eval(test, metric=Misclassification()) * 100))
p2 = [
b1,
Affine(nout=16, linear_name="b1_l1", **normrelu),
Affine(nout=10, linear_name="b1_l2", **normsigm)
]
p3 = [
b2,
Affine(nout=16, linear_name="b2_l1", **normrelu),
Affine(nout=10, linear_name="b2_l2", **normsigm)
]
# setup cost function as CrossEntropy
cost = Multicost(costs=[
GeneralizedCost(costfunc=CrossEntropyMulti()),
GeneralizedCost(costfunc=CrossEntropyBinary()),
GeneralizedCost(costfunc=CrossEntropyBinary())
],
weights=[1, 0., 0.])
# setup optimizer
optimizer = GradientDescentMomentum(0.1,
momentum_coef=0.9,
stochastic_round=args.rounding)
# initialize model object
alphas = [1, 0.25, 0.25]
mlp = Model(layers=Tree([p1, p2, p3], alphas=alphas))
# setup standard fit callbacks
decoder = [LookupTable(vocab_size=len(train_set.t_vocab), embedding_dim=embedding_dim,
init=init, name="LUT_de")]
decoder_connections = [] # link up recurrent layers
for ii in range(num_layers):
encoder.append(GRU(hidden_size, init, activation=Tanh(), gate_activation=Logistic(),
reset_cells=True, name="GRU1Enc"))
decoder.append(GRU(hidden_size, init, activation=Tanh(), gate_activation=Logistic(),
reset_cells=True, name="GRU1Dec"))
decoder_connections.append(ii)
decoder.append(Affine(train_set.nout, init, bias=init, activation=Softmax(), name="Affout"))
layers = Seq2Seq([encoder, decoder],
decoder_connections=decoder_connections,
name="Seq2Seq")
cost = GeneralizedCost(costfunc=CrossEntropyMulti(usebits=True))
model = Model(layers=layers)
optimizer = RMSProp(gradient_clip_value=gradient_clip_value, stochastic_round=args.rounding)
# configure callbacks
callbacks = Callbacks(model, eval_set=valid_set, **args.callback_args)
# train model
model.fit(train_set,
optimizer=optimizer,
num_epochs=args.epochs,
cost=cost, callbacks=callbacks)
# obtain predictions
def main():
# parse the command line arguments
parser = NeonArgparser(__doc__)
args = parser.parse_args()
logger = logging.getLogger()
logger.setLevel(args.log_thresh)
#Set up batch iterator for training images
train = ImgMaster(repo_dir='spectroDataTmp',
set_name='train',
inner_size=400,
subset_pct=100)
val = ImgMaster(repo_dir='spectroDataTmp',
set_name='validation',
inner_size=400,
subset_pct=100,
do_transforms=False)
test = ImgMaster(repo_dir='spectroTestDataTmp',
set_name='validation',
inner_size=400,
subset_pct=100,
do_transforms=False)
train.init_batch_provider()
test.init_batch_provider()
print "Constructing network..."
model = constuct_network()
model.load_weights(args.model_file)
#Optimizer
opt = Adadelta()
# configure callbacks
valmetric = TopKMisclassification(k=5)
callbacks = Callbacks(model,
train,
eval_set=val,
metric=valmetric,
**args.callback_args)
cost = GeneralizedCost(costfunc=CrossEntropyMulti())
#flag = input("Press Enter if you want to begin training process.")
print "Training network..."
print args.epochs
model.fit(train,
optimizer=opt,
num_epochs=args.epochs,
cost=cost,
callbacks=callbacks)
mets = model.eval(test, metric=valmetric)
print 'Validation set metrics:'
print 'LogLoss: %.2f, Accuracy: %.1f %%0 (Top-1), %.1f %% (Top-5)' % (
mets[0], (1.0 - mets[1]) * 100, (1.0 - mets[2]) * 100)
test.exit_batch_provider()
train.exit_batch_provider()
