def multiclass_model():
print('Building model...')
model = Sequential()
# 256 character-space (ascii only)
model.add(Embedding(128, embedding_size, input_length=window_size))
model.add(LSTM(2000, dropout=0.2, recurrent_dropout=0.2))
model.add(Dense(window_size, activation='sigmoid'))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['categorical_accuracy'])
print(ascii(model))
return model
def set_model(self, model):
with open('tmp_report.txt', 'w') as fh:
model.summary(print_fn=lambda x: fh.write(x + '\n'))
with open('tmp_report.txt', 'r') as f:
summary = f.read()
os.remove('tmp_report.txt')
self.send({"architecture":json.loads(model.to_json()),
"training": {
"optimizer":{**model.optimizer.get_config(), **{"name":model.optimizer.__class__.__name__.lower()}}
},
"ascii": ascii(model),
"summary": summary},
"model/")
def binary_model(embedding_size=105,
window_size=56,
window_step=4,
lstm_size=5480):
print('Building model...')
model = Sequential()
# 128 character-space (ascii only)
# best was lstm 2000, embedding 200
model.add(Embedding(128, embedding_size, input_length=window_size))
model.add(LSTM(lstm_size, dropout=0.2, recurrent_dropout=0.2))
model.add(Dense(1, activation='sigmoid'))
model.compile(
loss='binary_crossentropy',
optimizer='adam', #Adam(lr=0.001),
metrics=['binary_accuracy'])
print('-' * 20, 'Binary Model', '-' * 20)
print(ascii(model))
return model
batch_generator,
epochs=1, #epochs,
steps_per_epoch=N / batch_size,
# callbacks=[tbCallback, checkpointCallback]
)
# validate
larger_class, remove_items, N = precompute(filename=sys.argv[2],
multiclass=multiclass,
balance=not multiclass,
window_size=window_size,
window_step=window_step)
batch_generator = gen_training_data(filename=sys.argv[2],
multiclass=multiclass,
balance=False,
N=N,
window_size=window_size,
window_step=window_step,
batch_size=batch_size)
score, acc = model.evaluate_generator(batch_generator,
steps=N / batch_size)
name = modelname(embedding_size, lstm_size, acc, multiclass)
print('Saving Keras model', name)
model.save(os.path.abspath('.') + '/models/' + name)
print('\n', '+' * 20, 'Results', '+' * 20)
print(ascii(model))
print('Test score:', score)
print('Test accuracy:', acc)
from keras_diagram import ascii
def CNN(Bands,Size,Filepath):
model = Sequential()
model.add(Convolution2D(30, (15, 15), padding='valid', input_shape=(int(Bands), int(Size[0]), int(Size[1])), activation='relu', kernel_initializer='he_normal',, bias_initializer='zeros', name='convolution_1'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Convolution2D(30, (15, 15), activation='relu', kernel_initializer='he_normal'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Convolution2D(15, (5, 5), activation='relu', kernel_initializer='he_normal'))
model.add(Convolution2D(15, (5, 5), activation='relu', kernel_initializer='he_normal'))
#model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(512, activation='relu', kernel_initializer='he_normal'))
model.add(Dense(1, activation='sigmoid', kernel_initializer='he_normal'))
# Compile model
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
model.summary()
Map = ascii(model)
print('')
print(Map)
file = open('Output/'+Filepath+'/Model_Layout.txt','w')
file.write(Map)
file.close()
return model
def main():
"""
"""
train = pd.read_csv('data/train.csv')
test = pd.read_csv('data/test.csv')
X_train, X_val, y_train, y_val = train_test_split(train.drop(['label'],
axis=1),
train['label'],
test_size=VAL_SIZE,
stratify=train['label'],
random_state=SEED)
print(X_train.shape, X_val.shape, y_train.shape, y_val.shape)
train_dataset = X_train.values
train_labels = y_train
valid_dataset = X_val.values
valid_labels = y_val
train_dataset = train_dataset.reshape(X_train.shape[0], 28, 28, 1)
valid_dataset = valid_dataset.reshape(X_val.shape[0], 28, 28, 1)
train_dataset = train_dataset.astype('float32')
train_dataset /= 255
valid_dataset = valid_dataset.astype('float32')
valid_dataset /= 255
train_labels = np_utils.to_categorical(train_labels, 10)
valid_labels = np_utils.to_categorical(valid_labels, 10)
run_model = create_model()
run_model.summary()
start_time = datetime.datetime.now()
run_model.fit(train_dataset,
train_labels,
batch_size=BATCH_SIZE,
nb_epoch=NUM_EPOCHS,
verbose=1)
names = run_model.metrics_names
score = run_model.evaluate(valid_dataset, valid_labels, verbose=0)
print(names, score)
with open(logfile, "a") as myfile:
now = datetime.datetime.now()
results = zip(names, score)
summary = ascii(run_model)
print("\n{0} START\n{1} END".format(start_time, now), file=myfile)
print("learning rate: {} batch size: {} epochs: {} optimizer: adam".
format(LEARNING_RATE, BATCH_SIZE, NUM_EPOCHS),
file=myfile)
for name, score in results:
print("{0}: {1:.5f}".format(name, score), file=myfile)
print(summary, file=myfile)
test_dataset = test.values
test_dataset = test_dataset.reshape(test.shape[0], 28, 28, 1)
test_dataset = test_dataset.astype('float32')
test_dataset /= 255
predicted_labels = run_model.predict_classes(test_dataset)
generate_submission(predicted_labels)
input_L1 = Input(shape=(3, 32,32))
conv_L1 = Convolution2D(nb_filters,
nb_conv, nb_conv, init='glorot_uniform',
activation='relu', border_mode = 'same',
dim_ordering='th')
conved_L1 = conv_L1(input_L1)
mxpool_L1 = MaxPooling2D((2,2))(conved_L1)
conv_L2 = Convolution2D(nb_filters, nb_conv, nb_conv,
activation='relu', border_mode='same')
conved_L2 = conv_L1(mxpool_L1)
mxpool_L2 = MaxPooling2D((2,2))(conved_L2)
out_L3 = Flatten()(mxpool_L2)
output_L1 = Dense(128, activation='softmax')(out_L3)
output_L2 = Dense(11, activation='softmax')(output_L1)
model = Model(input_L1, output_L2)
model.summary()
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(X_train, y_train, batch_size=batch_size,validation_data=(X_test, y_test), nb_epoch=nb_epoch, verbose=1)
print(ascii(model))
def print_model(self):
print(ascii(self.model))
