def CNN_model(frameHeight, frameWidth):
model = Sequential()
model.add(Convolution2D(32, 3, 3, border_mode='same', init='he_normal', activation='relu',
input_shape=(1, int(frameHeight), int(frameWidth))))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.1))
model.add(Convolution2D(64, 3, 3, border_mode='same', init='he_normal', activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Convolution2D(128, 3, 3, border_mode='same', init='he_normal', activation='relu'))
model.add(MaxPooling2D(pool_size=(8, 8)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(32, W_regularizer=l2(1.26e-7)))
model.add(Activation('relu'))
model.add(Dense(2, W_regularizer=l2(1e-0)))
model.add(Activation('softmax'))
model.compile(Adam(lr=1e-3), loss='categorical_crossentropy')
plot(model, to_file='model.png')
return model
def cnn_optimise(W):
# Number of feature maps (outputs of convolutional layer)
N_fm = 300
# kernel size of convolutional layer
kernel_size = 8
conv_input_width = W.shape[1]
conv_input_height = 200 # maxlen of sentence
model = Sequential()
# Embedding layer (lookup table of trainable word vectors)
model.add(
Embedding(input_dim=W.shape[0], output_dim=W.shape[1], weights=[W], W_constraint=unitnorm(), init="uniform")
)
# Reshape word vectors from Embedding to tensor format suitable for Convolutional layer
model.add(Reshape(dims=(1, conv_input_height, conv_input_width)))
# first convolutional layer
model.add(Convolution2D(N_fm, kernel_size, conv_input_width, border_mode="valid", W_regularizer=l2(0.0001)))
# ReLU activation
model.add(Dropout(0.5))
model.add(Activation("relu"))
# aggregate data in every feature map to scalar using MAX operation
model.add(MaxPooling2D(pool_size=(conv_input_height - kernel_size + 1, 1), border_mode="valid"))
model.add(Dropout(0.5))
model.add(Flatten())
# Inner Product layer (as in regular neural network, but without non-linear activation function)
model.add(Dense(input_dim=N_fm, output_dim=1))
# SoftMax activation; actually, Dense+SoftMax works as Multinomial Logistic Regression
model.add(Activation("sigmoid"))
plot(model, to_file="./images/model.png")
return model
def make_keras_model(n_hidden_layers, dimof_middle, dimof_input):
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.utils import np_utils, generic_utils
from keras.wrappers.scikit_learn import KerasClassifier
dimof_output =1
print("dimof_input : ",dimof_input, "dimof_output : ", dimof_output)
batch_size = 100
dropout = 0.5
countof_epoch = 5
model = Sequential()
model.add(Dense(input_dim=dimof_input, output_dim=dimof_middle, init="glorot_uniform",activation='relu'))
model.add(Dropout(dropout))
for n in range(n_hidden_layers):
model.add(Dense(input_dim=dimof_middle, output_dim=dimof_middle, init="glorot_uniform",activation='relu'))
model.add(Dropout(dropout))
model.add(Dense(input_dim=dimof_middle, output_dim=dimof_output, init="glorot_uniform",activation='sigmoid'))
#Compiling (might take longer)
model.compile(class_mode='binary',loss='binary_crossentropy', optimizer='adam',metrics=["accuracy"])
visualize_util.plot(model, to_file='model.png')
return model
def visualize_model(model):
model.summary()
visualize_util.plot(model,
to_file='../../figures/%s.png' % model.name,
show_shapes=True,
show_layer_names=True)
def __init__(self, options):
"""Initialise network structure."""
# Pre-define batch for use in objective function feature_matching
self.timesteps = options['timesteps']
self.features = options['features']
self.gamma = options['gamma']
self.drop = options['dropout']
self.plot = options['plot']
self.epoch = options['epoch']
self.batch_size = options['batch_size']
self.init = options['layer_init']
self.pre_train_D = options['pre_train_D']
self.num_GRU = options["GRU_layers"]
self.is_mir_eval = options["IS_MIR_EVAL"]
self.G_lr = options["G_lr"]
self.D_lr = options["D_lr"]
self.GAN_lr = options["GAN_lr"]
self.mir_eval = {'SAR': [],
'SIR': [],
'SDR': []}
self.batch = np.zeros((self.batch_size, self.timesteps, self.features))
self.D__init__()
self.G__init__()
self.GAN__init__()
if self.plot:
plot(self.GAN, to_file='model.png')
def createmodel(img_channels,img_rows,img_cols,classes=1):
# This is a Sequential model. Graph models can be used in order to create more complex networks.
# Teaching Points:
# 1. Here we utilize the adam optimization algorithm. In order to use the SGD algorithm one could replace the {adam=keras.optimizers.Adadelta(lr=0)} line with {sgd = SGD(lr=0.0, momentum=0.9, decay=0.0, nesterov=False)} make sure you import the correct optimizer from keras.
# 2. This is a binary classification problem so make sure that the correct activation loss function combination is used. For such a problem the sigmoid activation function with the binary cross entropy loss is a good option
# 3. Since this is a binary problem use model.add(Dense(1)) NOT 2...
# 4. For multi class model this code can be easily modified by selecting the softmax as activation function and the categorical cross entropy as loss
model = Sequential()
# first set of CONV => RELU => POOL
model.add(Convolution2D(20, 5, 5, border_mode="same",
input_shape=(img_channels, img_rows, img_cols)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
# second set of CONV => RELU => POOL
model.add(Convolution2D(50, 5, 5, border_mode="same"))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
# set of FC => RELU layers
model.add(Flatten())
model.add(Dense(500))
model.add(Activation("relu"))
# softmax classifier
model.add(Dense(classes))
# model.add(Activation('relu'))
model.add(Activation('sigmoid'))
# learning schedule callback
adam=keras.optimizers.Adadelta(lr=0)
lrate = LearningRateScheduler(step_decay)
model.compile(loss='binary_crossentropy', optimizer=adam, metrics=['accuracy'])
plot(model, to_file='modellen.png')
return model
def evaluate_single_model(model, folder_name, model_name, features_test, labels_test, save_model=True):
cat_labels_test = np_utils.to_categorical(labels_test)
loss, acc = model.evaluate(features_test, cat_labels_test, verbose=2)
write = "**********Evaluating " + str(model_name) + "************\n"
write += 'Testing data size: ' + str(len(labels_test)) + '\n'
write += str(Counter(labels_test)) + '\n'
write += 'loss: ' + str(loss) + ' acc' + str("%.2f" % round(acc, 4)) + '\n'
result = model.predict(features_test)
result_label = np.argmax(result, 1)
gt_label = labels_test
con_matrix = confusion_matrix(gt_label, result_label)
write += str(con_matrix) + '\n'
write += "Classification report:\n"
write += str(classification_report(gt_label, result_label)) + '\n'
# create folder if not exists
if not os.path.exists(folder_name):
os.makedirs(folder_name)
if save_model:
model_json = model.to_json()
with open(folder_name + model_name + "_model.json", "w") as json_file:
json_file.write(model_json)
# serialize weights to HDF5
model.save_weights(folder_name + model_name + "_model.h5")
print("Saved model to disk")
plot(model, to_file=folder_name + model_name + "_model.png", show_shapes=True)
return write, acc
def imdb_test():
# set parameters:
max_features = 5000 # number of vocabulary
maxlen = 200 # padding
batch_size = 16
nb_epoch = 10
print('Loading data...')
(X_train, y_train), (X_test, y_test) = imdb.load_data(nb_words=max_features,
test_split=0.2)
print("Pad sequences (samples x time)")
X_train = sequence.pad_sequences(X_train, maxlen=maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=maxlen)
print('X_train shape:', X_train.shape)
print('X_test shape:', X_test.shape)
nb_classes = 2
y_train = np_utils.to_categorical(y_train, nb_classes)
y_test = np_utils.to_categorical(y_test, nb_classes)
model = imdb_cnn()
plot(model, to_file='./images/imdb_model.png')
# try using different optimizers and different optimizer configs
# model.compile(loss='binary_crossentropy', optimizer='adagrad', class_mode="binary")
model.compile(loss='categorical_crossentropy', optimizer='adagrad')
print("Train...")
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=nb_epoch, validation_data=(X_test, y_test),
show_accuracy=True, callbacks=[early_stopping])
score, acc = model.evaluate(X_test, y_test, batch_size=batch_size, show_accuracy=True)
print('Test score:', score)
print('Test accuracy:', acc)
def build_model(config):
sent_input = Input(shape=(config.max_len,),dtype='int32',name = "sent_input") # define inpute
sent_embedding = Embedding(config.vocab_size,config.word_dims,
# W_regularizer=l2(0.01),
dropout=0.2,
name="sent_w2v",)(sent_input)
convs = []
if config.pool_method=="mean":
pool_method = mean_1d
elif config.pool_method=="max":
pool_method = max_1d
for nb_filter,filter_length in zip(config.nb_filters,config.filter_lengths):
conv = Convolution1D(nb_filter=nb_filter,
filter_length=filter_length,
border_mode='valid',
activation='relu',
subsample_length=1,
name="conv_len_{}_nb_{}".format(filter_length,nb_filter))(sent_embedding)
pooling = Lambda(pool_method, output_shape=(nb_filter,),name="max_pooling_len_{}_nb_{}".format(filter_length,nb_filter))(conv)
convs.append(pooling)
concat = merge(convs, mode='concat',name = "sent_merge")
hidden = Dropout(p = config.hidden_dropout)(Dense(config.hidden_dims,activation="relu",name = "sent_hidden")(concat))
out = Activation('softmax')(Dense(config.label_size)(hidden))
model = Model(input=sent_input, output=out,name=config.model_name)
print("build completed")
model.compile(loss='sparse_categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print(model.summary())
from keras.utils.visualize_util import plot
plot(model, to_file='flatten_10.png',show_shapes=True)
return model
def compile_model(self, lr=0.0001, loss_weights=0.1):
optimizer = Adam(lr=lr)
loss = 'mse'
# loss = custom_objective
self.model.compile(optimizer=optimizer, loss=loss)
#metrics=['mse'])
plot(self.model, to_file='model.png')
def get_new_model(in_width, lstm_hidden, out_width, batch_size, ltype):
model = Sequential()
# if args.time_dist_dense:
# model.add(LSTM(in_width, lstm_hidden, return_sequences=True))
# model.add(TimeDistributedDense(lstm_hidden, out_width))
# else:
# if args.rnn:
# model.add(SimpleRNN(in_width, lstm_hidden, return_sequences=False))
# else:
print "Number of hidden layers:", args.depth
# model.add(LSTM(lstm_hidden , input_dim=in_width, return_sequences=False if args.depth == 0 else True))
for i in range(args.depth):
model.add(generate_layer(ltype, in_width, lstm_hidden, "last" if i+1 == args.depth else "first" if i==0 else "middle"))
# model.add(LSTM(lstm_hidden , input_dim=in_width, return_sequences=False if i+1 == args.depth else True))
model.add(Dropout(args.dropout))
model.add(Dense(out_width))
model.add(Activation('softmax'))
optimizer = Adagrad(clipnorm=args.clipnorm)
model.compile(loss='categorical_crossentropy', metrics=['accuracy'], optimizer=optimizer)
from keras.utils.visualize_util import plot
plot(model, show_shapes=True, to_file='model.png')
return model
def draw_model(model):
from IPython.display import SVG
from keras.utils.visualize_util import to_graph
from keras.utils.visualize_util import plot
SVG(to_graph(model).create(prog='dot', format='svg'))
plot(model, to_file='UFCNN/UFCNN_1.png')
def build_model(self):
lstm_branch = []
input_branch = []
for i in range(self.layers_number):
main_input = Input(shape=(self.lstm_timesteps, self.input_dim), name="main_input_" + str(i))
input_branch.append(main_input)
lstm_out = LSTM(self.lstm_hidden, return_sequences=True)(main_input)
# auxiliary_input = Input(batch_shape=(self.batch_size,1,self.lstm_timesteps), name='auxiliary_input'+str(i))
auxiliary_input = Input(shape=(1, self.lstm_timesteps), name="auxiliary_input" + str(i))
input_branch.append(auxiliary_input)
"""
x1 = Merge([lstm_out, auxiliary_input], mode=lambda x, y: (x*y).sum(axis=0),
name='merge_lstm_auxi'+str(i))
"""
x1 = merge([auxiliary_input, lstm_out], mode="dot", dot_axes=[2, 1], name="merge_lstm_auxi" + str(i))
assert x1
flatten = Reshape((self.lstm_hidden,))(x1)
c_input = Input(shape=(6,), name="c_input" + str(i))
input_branch.append(c_input)
x2 = merge([flatten, c_input], mode="concat")
x2 = Dense(
self.lstm_hidden, activation="relu", W_regularizer=l2(0.001), activity_regularizer=activity_l2(0.001)
)(x2)
assert x2
lstm_branch.append(x2)
lstm_all_out = merge(lstm_branch, mode="sum", name="lstm_all_out")
"""
dense_relu = Dense(self.lstm_hidden, activation='relu', W_regularizer=l2(0.001),
activity_regularizer=activity_l2(0.001))(lstm_all_out)
"""
final_loss = Dense(self.output_dim, name="main_output")(lstm_all_out)
self.model = Model(input_branch, output=final_loss)
self.model.compile(loss="mean_squared_error", optimizer="adagrad")
plot(self.model, to_file="multiple_model.png", show_shapes=True)
def save_model(model, name="model"):
model_json = model.to_json()
model_out = open(name + '.json', 'w')
model_out.write(model_json)
model_out.close()
plot(model, to_file='model_' + name + '.png')
model.save_weights('weights_' + name, overwrite=True)
def main():
#load data
#X_train,Y_train,X_valid,Y_valid,X_test=load_data(training_dir,valid_dir,test_dir,labels,sample)
#preprocess data by mean subtraction and normalization
#X_train,X_valid,X_test=preprocess(X_train,X_valid,X_test)
#del X_train
#del X_test
#or load pre-processed data from a previously saved hdf5 file:
data=h5py.File('imagenet.hdf5','r')
X_train=np.asarray(data['X_train'])
Y_train=np.asarray(data['Y_train'])
X_valid=np.asarray(data['X_valid'])
Y_valid=np.asarray(data['Y_valid'])
X_test=np.asarray(data['X_test'])
print(Y_valid.shape)
print(X_valid.shape)
#print "loaded data from pickle"
#OPTIONAL: save loaded/pre-processed data to a pickle to save time in the future
'''
print "saving preprocessed data to hdf5 file"
f=h5py.File('imagenet.hdf5','w')
dset_xtrain=f.create_dataset("X_train",data=X_train)
dset_ytrain=f.create_dataset("Y_train",data=Y_train)
dset_xvalid=f.create_dataset("X_valid",data=X_valid)
dset_yvalid=f.create_dataset("Y_valid",data=Y_valid)
dset_xtest=f.create_dataset("X_test",data=X_test)
f.flush()
f.close()
print "done saving pre-processed data to hdf5 file!"
'''
#train a VGG-like convent
vgg_model,history=vgg_train(X_train,Y_train)
train_scores=vgg_evaluate(vgg_model,X_train,Y_train)
print "VGG-like net training scores:"+str(train_scores)
valid_scores=vgg_evaluate(vgg_model,X_valid,Y_valid)
print "VGG-like net validation scores:"+str(valid_scores)
#Visualize the pretty model
plot(vgg_model,to_file="vgg_like_convnet.png")
predictions=vgg_model.predict(X_test,verbose=1)
class_predictions=vgg_model.predict_classes(X_test)
#save all the outputs!
sys.setrecursionlimit(50000)
output=open('vgg_like_results.pkl','w')
pickle.dump(history,output)
pickle.dump(train_scores,output)
pickle.dump(valid_scores,output)
pickle.dump(predictions,output)
pickle.dump(class_predictions,output)
output.close()
#train a Keras version of the ConvNet implemented in Assignment#2 in class
#TODO
#train AlexNet
'''
def train_and_predict():
imgs_train, imgs_output_train = load_train_data()
imgs_train = imgs_train.astype('float32')
mean = np.mean(imgs_train) # mean for data centering
std = np.std(imgs_train) # std for data normalization
imgs_train -= mean
imgs_train /= std
imgs_output_train = imgs_output_train.astype('float32')
imgs_output_train /= 255. # scale outputs to [0, 1]
imgs_test, imgs_output_test = load_test_data()
imgs_test = imgs_test.astype('float32')
imgs_test -= mean
imgs_test /= std
imgs_output_test = imgs_output_test.astype('float32')
imgs_output_test /= 255. # scale outputs to [0, 1]
print('-'*30)
print('Creating and compiling model...')
print('-'*30)
model = get_unet()
model_checkpoint = ModelCheckpoint('unet.hdf5', monitor='val_loss', save_best_only=True )
print('-'*30)
print('Fitting model...')
print('-'*30)
earlyStopping=EarlyStopping(monitor='val_loss', patience=20, verbose=0, mode='auto')
model.fit(imgs_train, imgs_output_train, batch_size=16, nb_epoch=100, verbose=1, callbacks=[model_checkpoint, earlyStopping], validation_split=0.1, validation_data=None, shuffle=True)
print('-'*30)
print('Loading saved weights...')
print('-'*30)
model.load_weights('unet.hdf5')
print('-'*30)
print('Predicting outputs on test data...')
print('-'*30)
imgs_output_test2 = model.predict(imgs_test, verbose=1)
np.save('imgs_output_test2.npy', imgs_output_test2)
print('-'*30)
print('Evaluating test loss...')
print('-'*30)
score = model.evaluate(imgs_test, imgs_output_test, batch_size=16)
print(score)
print('-'*30)
print('Ploting model...')
print('-'*30)
plot(model, to_file='model.png')
def plot_model(self, filename='brnn_model.png'):
"""
Plot model.
Arguments:
filename: {string}, the name/path to the file
to which the weights are going to be plotted.
"""
print "Plot Model %s ..." %filename
plot(self.model, to_file=filename)
def write_model_graph(self):
model = self.create_model()
# write dot text to file
with open(config["model_dot"], "wb") as out:
m = model_to_dot(model)
dot_text = m.create_dot()
out.write(dot_text)
print("Wrote .dot to {}".format(config["model_dot"]))
# write graph to file
plot(model, to_file=config["model_graph"], show_shapes=True)
def plot_model(self, filename):
"""
Plot model.
Arguments:
filename: {string}, the name/path to the file
to which the model graphic is plotted.
"""
print "Plot Model %s ..." %filename
plot(self.model, to_file=filename)
def draw_model(model):
from IPython.display import SVG
from keras.utils.visualize_util import to_graph
from keras.utils.visualize_util import plot
#graph = to_graph(model, show_shape=True)
#graph.write_png("UFCNN_1.png")
SVG(to_graph(model).create(prog='dot', format='svg'))
plot(model, to_file='UFCNN_1.png')
def build(self):
enc_size = self.size_of_env_observation()
argument_size = IntegerArguments.size_of_arguments
input_enc = InputLayer(batch_input_shape=(self.batch_size, enc_size), name='input_enc')
input_arg = InputLayer(batch_input_shape=(self.batch_size, argument_size), name='input_arg')
input_prg = Embedding(input_dim=PROGRAM_VEC_SIZE, output_dim=PROGRAM_KEY_VEC_SIZE, input_length=1,
batch_input_shape=(self.batch_size, 1))
f_enc = Sequential(name='f_enc')
f_enc.add(Merge([input_enc, input_arg], mode='concat'))
f_enc.add(MaxoutDense(128, nb_feature=4))
self.f_enc = f_enc
program_embedding = Sequential(name='program_embedding')
program_embedding.add(input_prg)
f_enc_convert = Sequential(name='f_enc_convert')
f_enc_convert.add(f_enc)
f_enc_convert.add(RepeatVector(1))
f_lstm = Sequential(name='f_lstm')
f_lstm.add(Merge([f_enc_convert, program_embedding], mode='concat'))
f_lstm.add(LSTM(256, return_sequences=False, stateful=True, W_regularizer=l2(0.0000001)))
f_lstm.add(Activation('relu', name='relu_lstm_1'))
f_lstm.add(RepeatVector(1))
f_lstm.add(LSTM(256, return_sequences=False, stateful=True, W_regularizer=l2(0.0000001)))
f_lstm.add(Activation('relu', name='relu_lstm_2'))
# plot(f_lstm, to_file='f_lstm.png', show_shapes=True)
f_end = Sequential(name='f_end')
f_end.add(f_lstm)
f_end.add(Dense(1, W_regularizer=l2(0.001)))
f_end.add(Activation('sigmoid', name='sigmoid_end'))
f_prog = Sequential(name='f_prog')
f_prog.add(f_lstm)
f_prog.add(Dense(PROGRAM_KEY_VEC_SIZE, activation="relu"))
f_prog.add(Dense(PROGRAM_VEC_SIZE, W_regularizer=l2(0.0001)))
f_prog.add(Activation('softmax', name='softmax_prog'))
# plot(f_prog, to_file='f_prog.png', show_shapes=True)
f_args = []
for ai in range(1, IntegerArguments.max_arg_num+1):
f_arg = Sequential(name='f_arg%s' % ai)
f_arg.add(f_lstm)
f_arg.add(Dense(IntegerArguments.depth, W_regularizer=l2(0.0001)))
f_arg.add(Activation('softmax', name='softmax_arg%s' % ai))
f_args.append(f_arg)
# plot(f_arg, to_file='f_arg.png', show_shapes=True)
self.model = Model([input_enc.input, input_arg.input, input_prg.input],
[f_end.output, f_prog.output] + [fa.output for fa in f_args],
name="npi")
self.compile_model()
plot(self.model, to_file='model.png', show_shapes=True)
def presave(self, model_directory):
"""
Save the model architecture to the given directory.
:param model_directory: Directory to save model and weights.
:return:
"""
try:
open(os.path.join(model_directory, self.__architecture_name), 'w').write(self.model.to_json(indent=2))
except Exception as e:
self.logger.warn("Model structure is not saved due to: %s" % repr(e))
plot(self.model, to_file=os.path.join(model_directory, self.__model_graph_name))
def visualize_model(args):
model = MoleculeVAE()
data, charset = load_dataset(args.data, split = False)
if os.path.isfile(args.model):
model.load(charset, args.model)
else:
raise ValueError("Model file %s doesn't exist" % args.model)
plot(model.autoencoder, to_file = args.outfile)
def main():
# init-----------------------------------------------------------------
train_samples = dtp.sample_paths_from_list(DATA_LOC, LIST_LOC+TRAIN_LIST)
eval_samples = dtp.sample_paths_from_list(DATA_LOC, LIST_LOC+EVAL_LIST)
classes = open(DICT, 'r+').read().splitlines()
nb_classes = len(classes)
batch_size = 180
# load pretrained models----------------------------------------------------
pretrained = load_model(PRETRAINED_LOC, PRETRAINED_NAME)
# generate RGB stream model------------------------------------------------------
if not model_exist(RGB_MODEL_NAME):
print 'Training RGB stream model...\n'
rgb_stream = eitel.create_single_stream(nb_classes, pretrained, tag='_rgb')
plot(rgb_stream, 'stream_model.png')
rgb_stream = train_model(rgb_stream, 0, batch_size, train_samples, eval_samples, classes)
if SAVE_MODEL:
save_model(rgb_stream, RGB_MODEL_NAME)
del rgb_stream
else:
print 'RGB stream model already exists...'
# generate RGB stream model------------------------------------------------------
if not model_exist(DEP_MODEL_NAME):
print 'Training depth stream model...\n'
dep_stream = eitel.create_single_stream(nb_classes, pretrained, tag='_dep')
dep_stream = train_model(dep_stream, 1, batch_size, train_samples, eval_samples, classes)
if SAVE_MODEL:
save_model(dep_stream, DEP_MODEL_NAME)
del dep_stream
else:
print 'Depth stream model already exists...'
# reload the model weights----------------------------------------------------
print 'Loading weights...'
rgb_stream = load_model(MODEL_LOC, RGB_MODEL_NAME)
dep_stream = load_model(MODEL_LOC, DEP_MODEL_NAME)
# fusion model-----------------------------------------------------------
print 'Fusing stream models...'
fusion_model = eitel.create_model_merge(rgb_stream, dep_stream, nb_classes)
del rgb_stream
del dep_stream
plot(fusion_model, 'fusion_model.png')
batch_size=200
fusion_model = train_model(fusion_model, 2, batch_size, train_samples, eval_samples, classes)
if SAVE_MODEL:
save_model(fusion_model, FUS_MODEL_NAME)
def train_network(X, Y, epochs=1, train=True, filename=None, load_model=False, finetune=False, task_id=None, independent=False):
global model, history
if load_model:
print('Loading model')
input_model = open(filename[0], 'r')
model = model_from_json(input_model.read())
input_model.close()
model.compile(optimizer='rmsprop', loss='categorical_crossentropy')
model.load_weights(filename[1])
print('Loaded(probably)')
if finetune:
lastLayer = None
print(model.layers[0].__dict__)
for layer in model.layers:
if (layer.name == 'dense_1'):
nn.conv = layer
lastLayer = layer
import h5py
if not load_model or finetune:
X, Y, _ = transform_data(X, Y, task_id)
print('Labels shape: ' + str(Y.shape))
print('Labels look like this : ')
print(Y[0])
if not finetune:
nn.construct_model(X[0].shape)
nn.add_new_task(len(Y[0]), independent)
model = nn.get_model().model
if train:
print('Training started')
input_model = open(filename[0], 'w')
input_model.write(model.to_json())
input_model.close()
# model.summary()
from keras.utils.visualize_util import plot
plot(model, to_file='model.png')
print('Fitting')
history = nn.get_model().fit(X, Y, epoch=epochs)
task_dims.append(Y.shape[1])
print('Training end')
if filename is not None:
input_model = open(filename[0], 'w')
input_model.write(model.to_json())
input_model.close()
model.save_weights(filename[1], overwrite=True)
def save_network(model, problem_name, operation_name, network_name):
""" saves the given network as a PNG image.
Args:
model: The Keras Suequential model.
problem_name: The name of target problem.
operation_name: The name of arithmetic operation.
network_name: The name of the network.
"""
model_name = problem_name + "_" + operation_name + network_name
plot(model, to_file=model_name + ".png", show_shapes=True)
def SA_sst():
((x_train_idx_data, y_train_valence, y_train_labels,
x_test_idx_data, y_test_valence, y_test_labels,
x_valid_idx_data, y_valid_valence, y_valid_labels,
x_train_polarity_idx_data, y_train_polarity,
x_test_polarity_idx_data, y_test_polarity,
x_valid_polarity_idx_data, y_valid_polarity), W) = nn_input() # build_keras_input_amended or build_keras_input
maxlen = 200 # cut texts after this number of words (among top max_features most common words)
batch_size = 16
(X_train, y_train), (X_test, y_test), (X_valid, y_valide) = (x_train_polarity_idx_data, y_train_polarity), (
x_test_polarity_idx_data, y_test_polarity), (x_valid_polarity_idx_data, y_valid_polarity)
print(len(X_train), 'train sequences')
print(len(X_test), 'test sequences')
# m= 0
# for i in X_train:
# if len(i) >0:
# for j in i:
# if j > m:
# m=j
# print(m)
max_features = W.shape[0] # shape of W: (13631, 300) , changed to 14027 through min_df = 3
print("Pad sequences (samples x time)")
X_train = sequence.pad_sequences(X_train, maxlen=maxlen)
X_test = sequence.pad_sequences(X_test, maxlen=maxlen)
X_valid = sequence.pad_sequences(X_valid, maxlen=maxlen)
print('X_train shape:', X_train.shape)
print('X_test shape:', X_test.shape)
nb_classes = 2
y_train = np_utils.to_categorical(y_train, nb_classes)
y_test = np_utils.to_categorical(y_test, nb_classes)
y_valide = np_utils.to_categorical(y_valide, nb_classes)
model = Deep_CNN(W)
plot(model, to_file='./images/model.png')
# try using different optimizers and different optimizer configs
# model.compile(loss='binary_crossentropy', optimizer='adagrad', class_mode="binary")
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='categorical_crossentropy', optimizer='adagrad') # adagrad
print("Train...")
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=20, validation_data=(X_test, y_test), show_accuracy=True,
callbacks=[early_stopping])
score, acc = model.evaluate(X_test, y_test, batch_size=batch_size, show_accuracy=True)
print('Test score:', score)
print('Test accuracy:', acc)
def check_print():
# Create the Model
x, img_input, CONCAT_AXIS, INP_SHAPE, DIM_ORDERING = create_model()
# Create a Keras Model - Functional API
model = Model(input=img_input,
output=[x])
model.summary()
# Save a PNG of the Model Build
plot(model, to_file='./Model/AlexNet.png')
model.compile(optimizer='rmsprop',
loss='categorical_crossentropy')
print('Model Compiled')
def createModel(settings, createPlot):
print("Building Model...")
model = Sequential()
# if (settings.depth > 1):
model.add(LSTM(settings.hiddenNodes, return_sequences=True, input_shape=(settings.sequence_size, settings.N_values)))
model.add(Dropout(settings.dropoutAmount))
# else:
# model.add(LSTM(settings.hiddenNodes, return_sequences=False, input_shape=(settings.sequence_size, settings.N_values)))
# inner_init='orthogonal'
# model.add(Dropout(settings.dropoutAmount))
for i in range(1, settings.depth-1):
model.add(LSTM(settings.hiddenNodes, return_sequences=True))
model.add(Dropout(settings.dropoutAmount))
model.add(LSTM(settings.hiddenNodes, return_sequences=False))
model.add(Dropout(settings.dropoutAmount))
# model.add(LSTM(settings.hiddenNodes, return_sequences=False, input_shape=(settings.sequence_size, settings.N_values)))
# model.add(LSTM(settings.hiddenNodes, return_sequences=True))
# model.add(LSTM(settings.hiddenNodes, return_sequences=True))
# model.add(LSTM(settings.hiddenNodes, return_sequences=True))
# model.add(LSTM(settings.hiddenNodes, return_sequences=False))
if (settings.l1Amount > 0):
model.add(Dense(settings.N_values, W_regularizer=l1(settings.l1Amount)))
elif settings.l2Amount > 0:
model.add(Dense(settings.N_values, W_regularizer=l2(settings.l2Amount)))
else:
model.add(Dense(settings.N_values))
model.add(Activation(settings.activation))
model.compile(loss=settings.lossType, optimizer=settings.optimizer)
settings.filename = settings.activation + "_" + settings.trainingset + "_nodes" + str(settings.hiddenNodes) +"_depth" + str(settings.depth) + "_seq" + str(settings.sequence_size) + "_drop" + str(settings.dropoutAmount) + "_L1r" + str(settings.l1Amount) + "_L2r" + str(settings.l2Amount)
save_model_scratch(model, settings.filename, 0, False)
with open('/var/scratch/jvranken/models/' + settings.filename + '/model_settings.txt', 'w') as settingsFile:
for (setting, value) in vars(settings).items():
settingsFile.write(setting + ': ' + str(value) + '\n')
if createPlot:
plot(model, to_file='/var/scratch/jvranken/models/' + settings.filename + '/model_layout.png', show_shapes=True)
return model
def train(x_train, y_train, x_test, y_test, vocab_size):
model = Sequential()
model.add(Embedding(vocab_size, 128, input_length=maxlen, dropout=0.5, mask_zero = True))
model.add(LSTM(128, dropout_W=0.5, dropout_U=0.1)) # try using a GRU instead, for fun
model.add(Dropout(0.5))
model.add(Dense(vocab_size))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop')
print('Train...')
batch_size = 32
plot(model, to_file='model.png', show_shape = True)
model.fit(x_train, y_train, batch_size=batch_size, nb_epoch=15,
validation_data=(x_test, y_test), show_accuracy=True)
score, acc = model.evaluate(x_test, y_test,
batch_size=batch_size,
show_accuracy=True)
print('Test score:', score)
print('Test accuracy:', acc)
dp2 = Dropout(0.5)(fc2)
output = Dense(1)(dp2)
model = Model(input=[uinput, iinput], output=output)
sgd = SGD(lr=0.05, decay=1e-6, momentum=0.9, nesterov=True)
# model.compile(optimizer=sgd,
# loss='mse')
model.compile(optimizer='adam', loss='mse')
model_name = 'mlp.hdf5'
model_checkpoint = ModelCheckpoint(path + model_name,
monitor='val_loss',
save_best_only=True)
plot(model,
to_file=path + '%s.png' % model_name.replace('.hdf5', ''),
show_shapes=True)
nb_epoch = 5
batch_size = 1024 * 6
load_model = False
if load_model:
model.load_weights(path + model_name)
model.fit([u_train, i_train],
y_train,
batch_size=batch_size,
nb_epoch=nb_epoch,
verbose=1,
shuffle=True,
x[1, :, :] -= 116.779
x[2, :, :] -= 123.68
else:
# 'RGB'->'BGR'
x = x[:, :, ::-1]
# Zero-center by mean pixel
x[:, :, 0] -= 103.939
x[:, :, 1] -= 116.779
x[:, :, 2] -= 123.68
return x
# create the base pre-trained model
base_model = VGG16(weights='imagenet')
plot(base_model,
to_file='./results/modelVGG16a.png',
show_shapes=True,
show_layer_names=True)
x = base_model.get_layer('fc2').output
x = base_model.layers[-2].output
# x = base_model.get_layer('block5_pool').output
# x = Flatten(name='flat')(x)
# x = Dense(4096, activation='relu', name='fc')(x)
x = Dense(8, activation='softmax', name='predictions')(x)
model = Model(input=base_model.input, output=x)
plot(model,
to_file='./results/modelVGG16b.png',
show_shapes=True,
show_layer_names=True)
for layer in base_model.layers:
def train(self,
max_sent_len,
train_inputs,
train_labels,
num_epochs=20,
mlp_size=1024,
mlp_activation='relu',
dropout=None,
embedding_file=None,
tune_embedding=True,
num_mlp_layers=2,
batch=None,
patience=5):
'''
train_inputs (list(numpy_array)): The two sentence inputs
train_labels (numpy_array): One-hot matrix indicating labels
num_epochs (int): Maximum number of epochs to run
mlp_size (int): Dimensionality of each layer in the MLP
dropout (dict(str->float)): Probabilities in Dropout layers after "embedding" and "encoder" (lstm)
embedding (numpy): Optional pretrained embedding
tune_embedding (bool): If pretrained embedding is given, tune it.
patience (int): Early stopping patience
'''
if dropout is None:
dropout = {}
num_label_types = train_labels.shape[
1] # train_labels is of shape (num_samples, num_label_types)
sent1_input_layer = Input(name='sent1',
shape=train_inputs[0].shape[1:],
dtype='int32')
sent2_input_layer = Input(name='sent2',
shape=train_inputs[1].shape[1:],
dtype='int32')
encoded_sent1, encoded_sent2 = self._get_encoded_sentence_variables(
max_sent_len,
sent1_input_layer,
sent2_input_layer,
dropout,
embedding_file,
tune_embedding,
batch=32 if batch == None else batch)
concat_sent_rep = merge([encoded_sent1, encoded_sent2], mode='concat')
mul_sent_rep = merge([encoded_sent1, encoded_sent2], mode='mul')
diff_sent_rep = merge([encoded_sent1, encoded_sent2],
mode=lambda l: l[0] - l[1],
output_shape=lambda l: l[0])
# Use heuristic from Mou et al. (2015) to get final merged representation
merged_sent_rep = merge([concat_sent_rep, mul_sent_rep, diff_sent_rep],
mode='concat')
current_projection = merged_sent_rep
for i in range(num_mlp_layers):
mlp_layer_i = Dense(output_dim=mlp_size,
activation=mlp_activation,
name="%s_layer_%d" % (mlp_activation, i))
current_projection = mlp_layer_i(current_projection)
if dropout is not None:
if "output" in dropout:
current_projection = Dropout(
dropout["output"])(current_projection)
softmax = Dense(output_dim=num_label_types,
activation='softmax',
name='softmax_layer')
label_probs = softmax(current_projection)
model = Model(input=[sent1_input_layer, sent2_input_layer],
output=label_probs)
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
self.model = model
print >> sys.stderr, "Entailment model summary:"
model.summary()
plot(model,
to_file='model_plot.png',
show_shapes=True,
show_layer_names=True)
best_accuracy = 0.0
num_worse_epochs = 0
for epoch_id in range(num_epochs):
print >> sys.stderr, "Epoch: %d" % epoch_id
history = model.fit(train_inputs,
train_labels,
validation_split=0.1,
nb_epoch=1)
validation_accuracy = history.history['val_acc'][
0] # history['val_acc'] is a list of size nb_epoch
if validation_accuracy > best_accuracy:
self.save_model(epoch_id)
self.best_epoch = epoch_id
num_worse_epochs = 0
best_accuracy = validation_accuracy
elif validation_accuracy < best_accuracy:
num_worse_epochs += 1
if num_worse_epochs >= patience:
print >> sys.stderr, "Stopping training."
break
self.save_best_model()
import argparse
from keras.models import model_from_json
from keras.utils.visualize_util import plot
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Network visualization')
parser.add_argument(
'model',
type=str,
help=
'Path to model definition json. Model weights should be on the same path.'
)
args = parser.parse_args()
with open(args.model, 'r') as jfile:
# NOTE: if you saved the file by calling json.dump(model.to_json(), ...)
# then you will have to call:
#
# model = model_from_json(json.loads(jfile.read()))\
#
# instead.
model = model_from_json(jfile.read())
model.compile("adam", "mse")
plot(model, show_shapes=True, to_file='img/model.png')
def main():
emotion_classifier = load_model(model_path)
emotion_classifier.summary()
plot(emotion_classifier, to_file='model.png')
model.add(Permute((2,1,3,4)))
print(model.layers[6].output_shape)
model.add(Reshape((100,16*5*4)))
model.add(LSTM(128,return_sequences=True))
model.add(LSTM(128,return_sequences=True))
model.add(Flatten())
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
model.summary()
modelname = 'cube_cnn_2l_tmr_filtered'
plot(model, to_file='/media/kong/9A8821088820E48B/Xcat/experiment/BCI2008/ds1a/model/arange_pole/'+modelname+'.png', show_shapes=True)
opt = Adam()
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=['accuracy'])
checkpointer = ModelCheckpoint(filepath='/media/kong/9A8821088820E48B/Xcat/experiment/BCI2008/ds1a/model/arange_pole/'+modelname+'.h5',verbose=1,save_best_only=True, monitor='val_acc')
model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
verbose=2, validation_data=(X_test, Y_test), callbacks=[checkpointer])
#model.save('/home/xcat/MasterPaper/model/' + modelname + '.h5')
score = model.evaluate(X_test, Y_test, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])
border_mode='valid',
input_shape=(1, img_rows, img_cols),
# W_regularizer = l1l2ld(l1 = 0., l2 = 0., ld = 0.),
W_regularizer=l2(l=0.),
b_regularizer=l2(l=0.)))
# model.add(AveragePooling2D(pool_size=(5, 5)))
# model.add(Dropout(0.25))
model.add(Flatten())
# model.add(Dense(128))
# model.add(Activation('relu'))
# model.add(Dropout(0.5))
model.add(Dense(nb_classes, W_regularizer=l2(l=0.), b_regularizer=l2(l=0.)))
model.add(Activation('softmax'))
plot(model, to_file="./mnist.png", show_shapes=True)
# sys.exit()
for i in range(len(model.layers)):
print("i: ", i)
print(model.layers[i].get_config())
# sys.exit()
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy'])
model.fit(X_train,
Y_train,
batch_size=batch_size,
def build_resnet_152(input_shape, num_outputs):
return ResNetBuilder.build(input_shape, num_outputs, bottleneck, [3, 8, 36, 3])
# Define resnet model between 18, 34, 50, 101 or 152
model = ResNetBuilder.build_resnet_18((3, img_width, img_height), 1)
print('-'*30)
print('Model summary...')
print('-'*30)
model.summary()
if(PRINT_MODEL):
print('-'*30)
print('Printing model...')
print('-'*30)
plot(model, to_file='resnet50_keras_model.png')
model.compile(loss='binary_crossentropy',
optimizer='sgd',
metrics=['accuracy'])
# this is the augmentation configuration we will use for training
train_datagen = ImageDataGenerator(
rescale=1./255,
shear_range=0,
rotation_range=40, # randomly rotate images in the range (degrees, 0 to 180)
width_shift_range=0.2, # randomly shift images horizontally (fraction of total width)
height_shift_range=0.2, # randomly shift images vertically (fraction of total height)
zoom_range=0.2,
horizontal_flip=True, # randomly flip images
vertical_flip=False) # randomly flip images
net['conv8_2_mbox_priorbox'], net['pool6_mbox_priorbox']
],
mode='concat',
concat_axis=1,
name='mbox_priorbox')
if hasattr(net['mbox_loc'], '_keras_shape'):
num_boxes = net['mbox_loc']._keras_shape[-1] // 4
elif hasattr(net['mbox_loc'], 'int_shape'):
num_boxes = K.int_shape(net['mbox_loc'])[-1] // 4
net['mbox_loc'] = Reshape((num_boxes, 4),
name='mbox_loc_final')(net['mbox_loc'])
net['mbox_conf'] = Reshape((num_boxes, num_classes),
name='mbox_conf_logits')(net['mbox_conf'])
net['mbox_conf'] = Activation('softmax',
name='mbox_conf_final')(net['mbox_conf'])
net['predictions'] = merge(
[net['mbox_loc'], net['mbox_conf'], net['mbox_priorbox']],
mode='concat',
concat_axis=2,
name='predictions')
model = Model(net['input'], net['predictions'])
return model
if __name__ == "__main__":
ssd = SSD512((512, 512, 3))
ssd.summary()
from keras.utils.visualize_util import plot
plot(ssd, to_file='ssd_model.png', show_layer_names=True, show_shapes=True)
question_net = Sequential()
question_net.add(q_embedding_layer)
# question_net.add(LSTM(EMBEDDING_DIM, consume_less='cpu'))
# question_net.add(RepeatVector(400))
# question_net.add(Dense(EMBEDDING_DIM))
# question_net.add(Bidirectional(GRU(EMBEDDING_DIM)))
# question_net.add(Activation('tanh'))
# question_net.add(Dropout(0.1))
question_net.add(Bidirectional(LSTM(EMBEDDING_DIM, return_sequences=True)))
# question_net.add(GRU(EMBEDDING_DIM, return_sequences=True))
# question_net.add(LSTM(EMBEDDING_DIM, return_sequences=True, consume_less='mem'))
print("question layer shape:", question_net.layers[-1].output_shape)
# question_net.add(Dense(1, activation='sigmoid'))
plot(question_net, to_file='question_net.png', show_shapes=True)
merged = Merge([passage_net, question_net], mode='dot')
# merged = Merge([passage_net, question_net], mode='cos')
print("merged layer shape:", question_net.layers[-1].output_shape)
model = Sequential()
model.add(merged)
# multiply passage by the dot product
# add softmax here
# model.add(Dropout(.5))
# model.add(TimeDistributedDense(100, activation='softmax'))
model.add(Permute((2, 1)))
print "-----------------"
print(hist.history.get('loss')[-1])
print "-----------------"
loss=hist.history.get('loss')
acc=hist.history.get('acc')
model.save('Share_Weight/model_122222.h5')
np.save('Share_Weight/loss_122222.npy',loss)
np.save('Share_Weight/acc_122222.npy',acc)
sendEmail('Modelling Done')
model=load_model('Share_Weight/model_122222.h5')
loss=np.load('Share_Weight/loss_122222.npy')
acc=np.load('Share_Weight/acc_122222.npy')
plot(model, to_file='Share_Weight/TreeModel_122222.png')
rangeNum=300000 #300000(start——end)组算例
avgerr=np.zeros((rangeNum,len(opL)))
preStart=200000
preNum=100000
print("predict")
inputnum = len(input[0, :])
outputnumL = len(outputL[0, :])
outputnumR = len(outputR[0, :])
predictend = preStart + preNum
# 真实数据作为测试输入
testinput = data[preStart:predictend, ip]
# 真实数据作为输出，用来对比效果
conv2 = Convolution2D(nb_filters, kernel_size_2[0], kernel_size_2[1], border_mode='same')
model.add(conv2)
model.add(Activation('relu'))
pool2 = MaxPooling2D(pool_size=pool_size_2)
model.add(pool2)
model.add(Flatten())
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
model.summary()
modelname = '2cnn_adam_max_relu'
plot(model, to_file='/home/xcat/experiment/BCI2008/ds1a/model/on-filtered-data/'+modelname+'.png', show_shapes=True)
opt = Adam()
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=['accuracy'])
checkpointer = ModelCheckpoint(filepath='/home/xcat/experiment/BCI2008/ds1a/model/on-filtered-data/'+modelname+'.h5',verbose=1,save_best_only=True, monitor='val_acc')
model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
verbose=2, validation_data=(X_test, Y_test), callbacks=[checkpointer])
#model.save('/home/xcat/MasterPaper/model/' + modelname + '.h5')
score = model.evaluate(X_test, Y_test, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])
def convNet_LeNet(
VERBOSE=1,
# normlize
NORMALIZE = True,
# Network Parameters
BATCH_SIZE = 128,
NUM_EPOCHS = 20,
# Number of convolutional filters
NUM_FILTERS = 32,
# side length of maxpooling square
NUM_POOL = 2,
# side length of convolution square
NUM_CONV = 3,
# dropout rate for regularization
DROPOUT_RATE = 0.5,
# hidden number of neurons first layer
NUM_HIDDEN = 128,
# validation data
VALIDATION_SPLIT=0.2, # 20%
# optimizer used
OPTIMIZER = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True),
# regularization
REGULARIZER = l2(0.01)
):
# Output classes, number of MINST DIGITS
NUM_CLASSES = 10
# Shape of an MINST digit image
SHAPE_X, SHAPE_Y = 28, 28
# Channels on MINST
IMG_CHANNELS = 1
# LOAD the MINST DATA split in training and test data
(X_train, Y_train), (X_test, Y_test) = mnist.load_data()
X_train = X_train.reshape(X_train.shape[0], 1, SHAPE_X, SHAPE_Y)
X_test = X_test.reshape(X_test.shape[0], 1, SHAPE_X, SHAPE_Y)
# convert in float32 representation for GPU computation
X_train = X_train.astype("float32")
X_test = X_test.astype("float32")
if (NORMALIZE):
# NORMALIZE each pixerl by dividing by max_value=255
X_train /= 255
X_test /= 255
print('X_train shape:', X_train.shape)
print(X_train.shape[0], 'train samples')
print(X_test.shape[0], 'test samples')
# KERAS needs to represent each output class into OHE representation
Y_train = np_utils.to_categorical(Y_train, NUM_CLASSES)
Y_test = np_utils.to_categorical(Y_test, NUM_CLASSES)
nn = Sequential()
#FIRST LAYER OF CONVNETS, POOLING, DROPOUT
# apply a NUM_CONV x NUM_CONF convolution with NUM_FILTERS output
# for the first layer it is also required to define the input shape
# activation function is rectified linear
nn.add(Convolution2D(NUM_FILTERS, NUM_CONV, NUM_CONV,
input_shape=(IMG_CHANNELS, SHAPE_X, SHAPE_Y) ))
nn.add(Activation('relu'))
nn.add(Convolution2D(NUM_FILTERS, NUM_CONV, NUM_CONV))
nn.add(Activation('relu'))
nn.add(MaxPooling2D(pool_size = (NUM_POOL, NUM_POOL)))
nn.add(Dropout(DROPOUT_RATE))
#SECOND LAYER OF CONVNETS, POOLING, DROPOUT
# apply a NUM_CONV x NUM_CONF convolution with NUM_FILTERS output
nn.add(Convolution2D( NUM_FILTERS, NUM_CONV, NUM_CONV))
nn.add(Activation('relu'))
nn.add(Convolution2D(NUM_FILTERS, NUM_CONV, NUM_CONV))
nn.add(Activation('relu'))
nn.add(MaxPooling2D(pool_size = (NUM_POOL, NUM_POOL) ))
nn.add(Dropout(DROPOUT_RATE))
# FLATTEN the shape for dense connections
nn.add(Flatten())
# FIRST HIDDEN LAYER OF DENSE NETWORK
nn.add(Dense(NUM_HIDDEN))
nn.add(Activation('relu'))
nn.add(Dropout(DROPOUT_RATE))
# OUTFUT LAYER with NUM_CLASSES OUTPUTS
# ACTIVATION IS SOFTMAX, REGULARIZATION IS L2
nn.add(Dense(NUM_CLASSES, W_regularizer=REGULARIZER))
nn.add(Activation('softmax') )
#summary
nn.summary()
#plot the model
plot(nn)
# set an early-stopping value
early_stopping = EarlyStopping(monitor='val_loss', patience=2)
# COMPILE THE MODEL
# loss_function is categorical_crossentropy
# optimizer is parametric
nn.compile(loss='categorical_crossentropy',
optimizer=OPTIMIZER, metrics=["accuracy"])
start = time.time()
# FIT THE MODEL WITH VALIDATION DATA
fitlog = nn.fit(X_train, Y_train, \
batch_size=BATCH_SIZE, nb_epoch=NUM_EPOCHS, \
verbose=VERBOSE, validation_split=VALIDATION_SPLIT, \
callbacks=[early_stopping])
elapsed = time.time() - start
# Test the network
results = nn.evaluate(X_test, Y_test, verbose=VERBOSE)
print('accuracy:', results[1])
# just to get the list of input parameters and their value
frame = inspect.currentframe()
args, _, _, values = inspect.getargvalues(frame)
# used for printing pretty arguments
print_Graph(fitlog, elapsed, args, values)
return fitlog
model.add(Flatten())
model.add(Dropout(0.1))
model.add(Dense(150))
model.add(Dense(50))
model.add(Dense(10))
model.add(Dense(1))
model.compile(loss='mse', optimizer='adam')
history_object = model.fit_generator(train_generator, samples_per_epoch = \
len(train_samples) * 6, validation_data=validation_generator, \
nb_val_samples=len(validation_samples) * 6, nb_epoch=5)
model.save('model.h5')
print(history_object.history.keys())
# plot the training and validation loss for each epoch
plt.plot(history_object.history['loss'])
plt.plot(history_object.history['val_loss'])
plt.title('model mean squared error loss')
plt.grid()
plt.ylabel('mean squared error loss')
plt.xlabel('epoch')
plt.legend(['training set', 'validation set'], loc='upper right')
plt.savefig('training.png')
plt.show()
# plot the CNN architecture
plot(model, to_file='model.png')
name="encode1")
encode1_tensor = encode1_layer(raw_input_tensor)
decode1_layer = Dense(output_dim=args.patch_length * 2,
activation='linear',
name="decode1")
decode1_tensor = decode1_layer(encode1_tensor)
# First model is input -> encode1 -> decode1:
autoencoder1 = Model(input=raw_input_tensor, output=decode1_tensor)
autoencoder1.compile(optimizer='adadelta', loss='mse')
if args.net_plot:
pngname = '%s/keras_autoencoder1.png' % (args.output_dir, )
epsname = '%s/keras_autoencoder1.eps' % (args.output_dir, )
print("Saving %s..." % (pngname, ))
plot(autoencoder1, to_file=pngname, show_shapes=True)
print("Saving %s..." % (epsname, ))
plot(autoencoder1, to_file=epsname, show_shapes=True)
if train_net:
# Train first autoencoder on raw input.
autoencoder1.fit(x_train,
x_train,
nb_epoch=500,
batch_size=256,
shuffle=True,
validation_data=(x_val, x_val))
# Stack the 2nd autoencoder (connecting to encode1):
encode2_layer = Dense(hidden2,
activation='sigmoid',
x = MaxPooling2D((2,2))(x)
for i in range(N):
x = conv3_block(x, k, dropout)
nb_conv += 2
x = AveragePooling2D((8,8))(x)
x = Flatten()(x)
x = Dense(nb_classes, activation='softmax')(x)
model = Model(ip, x)
if verbose: print("Wide Residual Network-%d-%d created." % (nb_conv, k))
return model
if __name__ == "__main__":
from keras.utils.visualize_util import plot
from keras.layers import Input
from keras.models import Model
init = (3, 32, 32)
wrn_28_10 = create_wide_residual_network(init, nb_classes=100, N=4, k=10, dropout=0.25)
model = Model(init, wrn_28_10)
model.summary()
plot(model, "WRN-28-10.png", show_shapes=True, show_layer_names=True)
type=str,
help='Path to model h5 file. Model should be on the same path.')
parser.add_argument(
'image_folder',
type=str,
nargs='?',
default='',
help=
'Path to image folder. This is where the images from the run will be saved.'
)
args = parser.parse_args()
# check that model Keras version is same as local Keras version
f = h5py.File(args.model, mode='r')
model_version = f.attrs.get('keras_version')
keras_version = str(keras_version).encode('utf8')
if model_version != keras_version:
print('You are using Keras version ', keras_version,
', but the model was built using ', model_version)
model = load_model(args.model)
if not pydot.find_graphiz():
raise ImportError('Install pydot and graphviz.')
else:
plot(model,
to_file='model.png',
show_shapes=True,
show_layer_names=False)
x=Activation('relu')(x)
seq_output=Flatten()(x)
print('INFO - %s'%('building concatenate model.'))
units=512
x=Merge(mode='concat',concat_axis=1)([reg_output,seq_output])
x=Dense(units,activation='relu')(x)
x=Dropout(0.5,seed=42)(x)
x=Dense(units,activation='relu')(x)
x=Dropout(0.5,seed=42)(x)
rgs_output=Dense(1,activation='linear',name='rgs_output')(x)
model=Model(input=[reg_input,seq_input],output=[rgs_output])
model.compile(loss={'rgs_output':'mean_squared_error'},optimizer='sgd')
plot(model, show_shapes=True,to_file='%s/model.eps'%(fig_dir))
print('INFO - %s'%('loading data.'))
train,val,test=input_data.read_data_sets(train_pct=80,val_pct=10,test_pct=10)
print('INFO - %s'%('training model.'))
reduce_lr=ReduceLROnPlateau(verbose=1,factor=0.5, patience=5)
early_stopping=EarlyStopping(monitor='val_loss',patience=10)
checkpoint=ModelCheckpoint(filepath="%s/model.{epoch:02d}-{val_loss:.4f}.hdf5"%(log_dir), monitor='val_loss')
batchhistory=BatchHistory(val_data=val,loss_function='mse',every_n_batch=1000)
history=model.fit({'seq_input':train['seq'],'reg_input':train['reg']},{'rgs_output':train['expr']},
validation_data=({'seq_input':val['seq'],'reg_input':val['reg']},{'rgs_output':val['expr']}),
nb_epoch=100,
batch_size=100,
callbacks=[early_stopping,checkpoint,reduce_lr,batchhistory],
def train():
X = np.load('X_train.npy')
y = np.load('y_train.npy')
X = np.reshape(X, (30, 512, 512, 1))
# tf dim_ordering
y = np.reshape(y, (30, 512, 512, 1))
# tf dim_ordering
# Extend radius
radius = 64
X = np.pad(X, ((0, 0), (radius, radius), (radius, radius), (0, 0)),
'reflect')
# tf dim_ordering
y = np.pad(y, ((0, 0), (radius, radius), (radius, radius), (0, 0)),
'reflect')
# tf dim_ordering
# Preprocess the label
y = y / 255
y = y.astype('float32')
X = X.astype('float32')
# Merging 7 times
X0 = X
y0 = y
X1 = X[::1, ::1, ::-1, ::1]
y1 = y[::1, ::1, ::-1, ::1]
X2 = X[::1, ::-1, ::1, ::1]
y2 = y[::1, ::-1, ::1, ::1]
X3 = np.transpose(X, (0, 2, 1, 3))
y3 = np.transpose(y, (0, 2, 1, 3))
X4 = scipy.ndimage.interpolation.rotate(X, angle=90, axes=(1, 2))
y4 = scipy.ndimage.interpolation.rotate(y, angle=90, axes=(1, 2))
X5 = scipy.ndimage.interpolation.rotate(X, angle=180, axes=(1, 2))
y5 = scipy.ndimage.interpolation.rotate(y, angle=180, axes=(1, 2))
X6 = scipy.ndimage.interpolation.rotate(X, angle=270, axes=(1, 2))
y6 = scipy.ndimage.interpolation.rotate(y, angle=270, axes=(1, 2))
X = np.concatenate((X0, X1, X2, X3, X4, X5, X6), axis=0)
y = np.concatenate((y0, y1, y2, y3, y4, y5, y6), axis=0)
##########
print("Y median", np.median(y))
print("X shape", X.shape)
print("X dtype", X.dtype)
print("Y shape", y.shape)
print("Y dtype", y.dtype)
nb_iter = 1001
epochs_per_iter = 2
batch_size = 1
print('-' * 30)
print('Creating and compiling model...')
print('-' * 30)
model = get_uplusnet()
model.reset_states()
# model.load_weights("model_300.hdf5")
# graph = to_graph(model, show_shape=True)
# graph.write_png("model.png")
plot(model, to_file='model.png', show_shapes=True)
nb_folds = 3
# kfolds = KFold(len(y), nb_folds)
kfolds = KFold(nb_folds)
# Perform cross validation on the data
for iter in range(nb_iter):
print('-' * 50)
print('Iteration {0}/{1}'.format(iter + 1, nb_iter))
print('-' * 50)
# Shuffle the data
print('Shuffle data...')
seed = np.random.randint(1, 10e6)
np.random.seed(seed)
np.random.shuffle(X)
np.random.seed(seed)
np.random.shuffle(y)
f = 0
for train, valid in kfolds.split(X):
# for train, valid in kfolds:
print('=' * 50)
print('Fold', f + 1)
f += 1
# Extract train, validation set
X_train = X[train]
X_valid = X[valid]
y_train = y[train]
y_valid = y[valid]
print('Augmenting data for training...')
X_train, y_train = augment_data(
X_train, y_train) # Data augmentation for training
X_valid, y_valid = augment_data(
X_valid, y_valid) # Data augmentation for training
print("X_train", X_train.shape)
print("y_train", y_train.shape)
# Normalize
# X_train = X_train/255.0
# X_valid = X_valid/255.0
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(
featurewise_center=False,
samplewise_center=True,
featurewise_std_normalization=False,
samplewise_std_normalization=True,
rescale=1 / 255.0)
train_datagen.fit(X_train)
valid_datagen = ImageDataGenerator(
featurewise_center=False,
samplewise_center=True,
featurewise_std_normalization=False,
samplewise_std_normalization=True,
rescale=1 / 255.0)
valid_datagen.fit(X_valid)
# checkpoint the best model
filepath = "model-best.hdf5"
checkpoint = ModelCheckpoint(filepath,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min')
callbacks_list = [
TensorBoard(
log_dir='/home/Pearl/quantm/isbi12_210_keras_uplus/logs/')
]
# datagen = ImageDataGenerator(
# samplewise_center=True,
# samplewise_std_normalization=True,
# zca_whitening=True)
history = model.fit(X_train,
y_train,
verbose=1,
shuffle=True,
nb_epoch=epochs_per_iter,
batch_size=batch_size,
callbacks=callbacks_list,
validation_data=(X_valid, y_valid))
# history = model.fit_generator(
# train_datagen.flow(X_train, y_train, batch_size=batch_size),
# verbose=1,
# samples_per_epoch=len(X_train),
# nb_epoch=epochs_per_iter,
# callbacks=callbacks_list,
# validation_data=valid_datagen.flow(X_valid, y_valid, batch_size=batch_size),
# nb_val_samples=len(X_valid)
# )
# list all data in history
# print(history.history.keys())
# Plot
# # summarize history for accuracy
# plt.plot(history.history['acc'])
# plt.plot(history.history['val_acc'])
# plt.title('model accuracy')
# plt.ylabel('accuracy')
# plt.xlabel('epoch')
# plt.legend(['train', 'test'], loc='upper left')
# plt.show()
# summarize history for loss
# plt.plot(history.history['loss'])
# plt.plot(history.history['val_loss'])
# plt.title('model loss')
# plt.ylabel('loss')
# plt.xlabel('epoch')
# plt.legend(['train', 'test'], loc='upper left')
# plt.show()
if iter % 10 == 0:
fname = 'models/model_%03d.hdf5' % (iter)
model.save_weights(fname, overwrite=True)
from keras.applications.vgg16 import VGG16
from keras.preprocessing import image
from keras.applications.vgg19 import preprocess_input
from keras.models import Model
from keras import backend as K
from keras.utils.visualize_util import plot
import numpy as np
import matplotlib.pyplot as plt
#load VGG model
base_model = VGG16(weights='imagenet')
#visalize topology in an image
plot(base_model, to_file='modelVGG16.png', show_shapes=True, show_layer_names=True)
#read and process image
img_path = '/data/MIT/test/coast/art1130.jpg'
img = image.load_img(img_path, target_size=(224, 224))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
#crop the model up to a certain layer
model = Model(input=base_model.input, output=base_model.get_layer('block5_conv2').output)
#get the features from images
features = model.predict(x)
if K.image_dim_ordering() == 'th':
#theano and thensorflow deal with tensor in different order
weights = base_model.get_layer('block1_conv1').get_weights()
aux_out = Flatten()(aux_out)
aux_out = Dense(nb_classes, activation='softmax')(aux_out)
# Reduction Resnet B
x = reduction_resnet_B(x)
# 5 x Inception Resnet C
for i in range(5):
x = inception_resnet_C(x, scale_residual=scale)
# Average Pooling
x = AveragePooling2D((8,8))(x)
# Dropout
x = Dropout(0.8)(x)
x = Flatten()(x)
# Output
out = Dense(output_dim=nb_classes, activation='softmax')(x)
model = Model(init, output=[out, aux_out], name='Inception-Resnet-v1')
return model
if __name__ == "__main__":
from keras.utils.visualize_util import plot
inception_resnet_v1 = create_inception_resnet_v1()
plot(inception_resnet_v1, to_file="Inception ResNet-v1.png", show_shapes=True)
logger.info('Initializing lookup table')
emb_reader = EmbReader(emb_path, emb_dim=emb_dim)
model_layer_index = 0
for test in model.layers:
if (test.name == 'SplusEmbedding' or test.name == 'SminusEmbedding'
or test.name == 'TplusEmbedding'
or test.name == 'TminusEmbedding'):
model.layers[model_layer_index].W.set_value(
emb_reader.get_emb_matrix_given_vocab(
vocab, model.layers[model_layer_index].W.get_value()))
model_layer_index += 1
model.compile(loss=loss, optimizer=optimizer, metrics=[metric])
plot(model, to_file=out_dir + '/model.png')
logger.info(' Done')
###############################################################################################################################
## Save model architecture
#
logger.info('Saving model architecture')
with open(out_dir + '/model_arch.json', 'w') as arch:
arch.write(model.to_json(indent=2))
logger.info(
'---------------------------------------------------------------------------------------'
)
if args.train is not None:
do_training = True
if args.model is not None:
load_model = True
model_to_load = args.model
if args.epochs is not None:
nb_epoch = args.epochs
print("command line arguments")
print(args)
train_indices, validation_indices, test_indices, labels, filenames = load_data(fail_data, pass_data)
model = qc_model()
model.summary()
plot(model, to_file="model.png")
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=4, min_lr=0.001)
stop_early = EarlyStopping(monitor='val_loss', min_delta=0, patience=5, verbose=0, mode='auto')
model_checkpoint = ModelCheckpoint(images_dir + "models/best_model.hdf5", monitor="val_acc", verbose=0, save_best_only=True, save_weights_only=False, mode='auto')
hist = model.fit_generator(batch(train_indices, labels, 2,True), nb_epoch=400, samples_per_epoch=len(train_indices), validation_data=batch(validation_indices, labels, 2), nb_val_samples=len(validation_indices), callbacks=[model_checkpoint], class_weight = {0:.7, 1:.3})
model.load_weights(images_dir + 'models/best_model.hdf5')
test_scores = []
sensitivities = []
specificities = []
import numpy as np
from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.utils.visualize_util import plot
model = Sequential()
model.add(Dense(32, input_dim=500))
model.add(Activation(activation='sigmoid'))
model.add(Dense(1))
model.add(Activation(activation='sigmoid'))
model.compile(optimizer='rmsprop', loss='mse', metrics=['mean_squared_error'])
data = np.random.random((1000, 500))
labels = np.random.randint(2, size=(1000, 1))
score = model.evaluate(data, labels, verbose=0)
print "Before Training:", zip(model.metrics_names, score)
model.fit(data, labels, nb_epoch=10, batch_size=32, verbose=0)
score = model.evaluate(data, labels, verbose=0)
print "After Training:", zip(model.metrics_names, score)
plot(model, to_file='s4.png', show_shapes=True)
# Before Training: [('loss', 0.26870122766494753), ('mean_squared_error', 0.26870122766494753)]
# After Training: [('loss', 0.22180086207389832), ('mean_squared_error', 0.22180086207389832)]
from keras.utils.visualize_util import plot
from keras.models import load_model
import argparse
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Network Diagram')
parser.add_argument('-m', '--model', type=str, help='Path to h5 model file.')
args = parser.parse_args()
net = load_model(args.model)
print("Writing out 'model.png'.")
plot(net, to_file='model.png')
model.add(pool2)
model.add(Permute((2, 1, 3)))
model.add(Reshape((56, 40 * 56)))
model.add(LSTM(128, return_sequences=True))
model.add(LSTM(128))
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
model.summary()
modelname = 'image-method'
plot(model,
to_file='/home/xcat/experiment/ImageNet/model/' + modelname + '.png',
show_shapes=True)
opt = Adam()
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=['accuracy'])
checkpointer = ModelCheckpoint(
filepath='/home/xcat/experiment/ImageNet/model/' + modelname + 'h5',
verbose=1,
save_best_only=True,
monitor='val_acc')
model.fit(X_train,
Y_train,
batch_size=batch_size,
def run_cifar10(batch_size,
nb_epoch,
depth,
nb_dense_block,
nb_filter,
growth_rate,
dropout_rate,
learning_rate,
weight_decay,
plot_architecture):
""" Run CIFAR10 experiments
:param batch_size: int -- batch size
:param nb_epoch: int -- number of training epochs
:param depth: int -- network depth
:param nb_dense_block: int -- number of dense blocks
:param nb_filter: int -- initial number of conv filter
:param growth_rate: int -- number of new filters added by conv layers
:param dropout_rate: float -- dropout rate
:param learning_rate: float -- learning rate
:param weight_decay: float -- weight decay
:param plot_architecture: bool -- whether to plot network architecture
"""
###################
# Data processing #
###################
# the data, shuffled and split between train and test sets
(X_train, y_train), (X_test, y_test) = cifar10.load_data()
nb_classes = len(np.unique(y_train))
img_dim = X_train.shape[1:]
if K.image_dim_ordering() == "th":
n_channels = X_train.shape[1]
else:
n_channels = X_train.shape[-1]
# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
# Normalisation
X = np.vstack((X_train, X_test))
# 2 cases depending on the image ordering
if K.image_dim_ordering() == "th":
for i in range(n_channels):
mean = np.mean(X[:, i, :, :])
std = np.std(X[:, i, :, :])
X_train[:, i, :, :] = (X_train[:, i, :, :] - mean) / std
X_test[:, i, :, :] = (X_test[:, i, :, :] - mean) / std
elif K.image_dim_ordering() == "tf":
for i in range(n_channels):
mean = np.mean(X[:, :, :, i])
std = np.std(X[:, :, :, i])
X_train[:, :, :, i] = (X_train[:, :, :, i] - mean) / std
X_test[:, :, :, i] = (X_test[:, :, :, i] - mean) / std
###################
# Construct model #
###################
model = densenet.DenseNet(nb_classes,
img_dim,
depth,
nb_dense_block,
growth_rate,
nb_filter,
dropout_rate=dropout_rate,
weight_decay=weight_decay)
# Model output
model.summary()
# Build optimizer
opt = Adam(lr=learning_rate, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=["accuracy"])
if plot_architecture:
from keras.utils.visualize_util import plot
plot(model, to_file='./figures/densenet_archi.png', show_shapes=True)
####################
# Network training #
####################
print("Training")
list_train_loss = []
list_test_loss = []
list_learning_rate = []
for e in range(nb_epoch):
if e == int(0.5 * nb_epoch):
K.set_value(model.optimizer.lr, np.float32(learning_rate / 10.))
if e == int(0.75 * nb_epoch):
K.set_value(model.optimizer.lr, np.float32(learning_rate / 100.))
split_size = batch_size
num_splits = X_train.shape[0] / split_size
arr_splits = np.array_split(np.arange(X_train.shape[0]), num_splits)
l_train_loss = []
start = time.time()
for batch_idx in arr_splits[:10]:
X_batch, Y_batch = X_train[batch_idx], Y_train[batch_idx]
train_logloss, train_acc = model.train_on_batch(X_batch, Y_batch)
l_train_loss.append([train_logloss, train_acc])
test_logloss, test_acc = model.evaluate(X_test,
Y_test,
verbose=0,
batch_size=64)
list_train_loss.append(np.mean(np.array(l_train_loss), 0).tolist())
list_test_loss.append([test_logloss, test_acc])
list_learning_rate.append(float(K.get_value(model.optimizer.lr)))
# to convert numpy array to json serializable
print('Epoch %s/%s, Time: %s' % (e + 1, nb_epoch, time.time() - start))
d_log = {}
d_log["batch_size"] = batch_size
d_log["nb_epoch"] = nb_epoch
d_log["optimizer"] = opt.get_config()
d_log["train_loss"] = list_train_loss
d_log["test_loss"] = list_test_loss
d_log["learning_rate"] = list_learning_rate
json_file = os.path.join('./log/experiment_log_cifar10.json')
with open(json_file, 'w') as fp:
json.dump(d_log, fp, indent=4, sort_keys=True)
def train():
#
# SaveModel is a CallBack class that we can use to save the model for each epoch
# This allows us to easily test each epoch on the simulator. The simulator seems like
# a better validation than just the validation data set
class SaveModel(Callback):
def on_epoch_end(self, epoch, logs={}):
epoch += 1
if (epoch > 0):
#with open ('model-' + str(epoch) + '.json', 'w') as file:
# file.write (model.to_json ())
# file.close ()
#model.save_weights ('model-' + str(epoch) + '.h5')
model.save('model-' + str(epoch) + '.h5')
#
# load the model
#
model = get_model()
# Keras has a nice tool to create an image of our network
from keras.utils.visualize_util import plot
plot(model, to_file='car_model.png', show_shapes=True)
print("Loaded model")
# load the data
xs, ys = data.loadTraining()
# split the dataset into training and validation 80% / 20%
train_xs = xs[:int(len(xs) * 0.8)]
train_ys = ys[:int(len(xs) * 0.8)]
val_xs = xs[-int(len(xs) * 0.2):]
val_ys = ys[-int(len(xs) * 0.2):]
# load the validation dataset, it is better not generate an image each time - process them once
# Use the validation process function, it doesn't augment the image, just resizes it
X, y = data.getValidationDataset(val_xs, val_ys,
data.processImageValidation)
print(model.summary())
print("Loaded validation datasetset")
print("Total of", len(train_ys))
print("Training..")
checkpoint_path = "weights.{epoch:02d}-{val_loss:.2f}.hdf5"
checkpoint = ModelCheckpoint(checkpoint_path,
verbose=1,
save_best_only=False,
save_weights_only=False,
mode='auto')
# I tried using the earlystopping callback, but now I run it for a fixed number of epochs and test to see which is best
earlystopping = EarlyStopping(monitor='val_loss',
min_delta=0,
patience=2,
verbose=1,
mode='auto')
res = model.fit_generator(data.generator(train_xs, train_ys, 256),
validation_data=(X, y),
samples_per_epoch=100 * 256,
nb_epoch=epochs,
verbose=1,
callbacks=[SaveModel()])
# pickle and dump the history so we can graph it in a notebook
history = res.history
with open('history.p', 'wb') as f:
pickle.dump(history, f)
# Set the calculated weights and biases
model.get_layer('h1').set_weights(W1b1)
model.get_layer('h2').set_weights(W2b2)
model.get_layer('h3').set_weights(W3b3)
model.get_layer('h4').set_weights(W4b4)
model.get_layer('h5').set_weights(W5b5)
model.get_layer('h6').set_weights(W6b6)
model.get_layer('out').set_weights(Wobo)
# Compile model
current_dir = os.path.dirname(os.path.realpath(__file__))
model_path = os.path.join(current_dir, "FullyConnectedNetworkPrelu.png")
plot(model, to_file=model_path, show_shapes=True)
adam=keras.optimizers.Adam(lr=0.0005, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model.compile(loss='binary_crossentropy', optimizer=adam,metrics=['categorical_accuracy'])
# learning schedule callback
history=History()
lrate = LearningRateScheduler(step_decay)
callbacks_list = [lrate,history]
#model Fitting
print "Training..."
model.fit(X_train, y_train,validation_data=(X_test,y_test),nb_epoch=2000, batch_size=X_train.shape[0], callbacks=callbacks_list, verbose=1)
#model.fit(X_train, y_train,validation_data=(X_test,y_test),nb_epoch=550, batch_size=X_train.shape[0],class_weight={0:1, 1:6756.0/271}, callbacks=callbacks_list, verbose=1)
#Model prediction
predicted=model.predict_proba(X_test,batch_size=25)
def make_one_net_model(self, cf, in_shape, loss, metrics, optimizer):
# Create the *Keras* model
if cf.model_name == 'fcn8':
model = build_fcn8(
in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
#path_weights='weights/pascal-fcn8s-dag.mat')
path_weights=None)
elif cf.model_name == 'unet':
model = build_unet(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
path_weights=None)
elif cf.model_name == 'segnet_basic':
model = build_segnet(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
path_weights=None,
basic=True)
elif cf.model_name == 'segnet_vgg':
model = build_segnet(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
path_weights=None,
basic=False)
elif cf.model_name == 'resnetFCN':
model = build_resnetFCN(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
path_weights=None)
elif cf.model_name == 'densenetFCN':
model = build_densenetFCN(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
freeze_layers_from=cf.freeze_layers_from,
path_weights=None)
elif cf.model_name == 'lenet':
model = build_lenet(in_shape, cf.dataset.n_classes,
cf.weight_decay)
elif cf.model_name == 'alexNet':
model = build_alexNet(in_shape, cf.dataset.n_classes,
cf.weight_decay)
elif cf.model_name == 'vgg16':
model = build_vgg(in_shape,
cf.dataset.n_classes,
16,
cf.weight_decay,
load_pretrained=cf.load_imageNet,
freeze_layers_from=cf.freeze_layers_from)
elif cf.model_name == 'vgg19':
model = build_vgg(in_shape,
cf.dataset.n_classes,
19,
cf.weight_decay,
load_pretrained=cf.load_imageNet,
freeze_layers_from=cf.freeze_layers_from)
elif cf.model_name == 'resnet50':
model = build_resnet50(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
load_pretrained=cf.load_imageNet,
freeze_layers_from=cf.freeze_layers_from)
elif cf.model_name == 'InceptionV3':
model = build_inceptionV3(in_shape,
cf.dataset.n_classes,
cf.weight_decay,
load_pretrained=cf.load_imageNet,
freeze_layers_from=cf.freeze_layers_from)
elif cf.model_name == 'yolo':
model = build_yolo(in_shape,
cf.dataset.n_classes,
cf.dataset.n_priors,
load_pretrained=cf.load_imageNet,
freeze_layers_from=cf.freeze_layers_from)
else:
raise ValueError('Unknown model')
# Load pretrained weights
if cf.load_pretrained:
print(' loading model weights from: ' + cf.weights_file + '...')
model.load_weights(cf.weights_file, by_name=True)
# Compile model
model.compile(loss=loss, metrics=metrics, optimizer=optimizer)
# Show model structure
if cf.show_model:
model.summary()
plot(model, to_file=os.path.join(cf.savepath, 'model.png'))
# Output the model
print(' Model: ' + cf.model_name)
# model is a keras model, Model is a class wrapper so that we can have
# other models (like GANs) made of a pair of keras models, with their
# own ways to train, test and predict
return One_Net_Model(model, cf, optimizer)
