def __init__(self):
self.model = None
self.epochs_trained = 0
# 读取或建立模型
if os.path.exists(Config.model_save_path):
self.model = keras.models.load_model(Config.model_save_path)
# 读取已训练的epochs数目
with open(Config.model_epochs_trained_path) as f:
line = f.readlines()[0]
self.epochs_trained = int(line)
else:
self.build_model()
self.model.summary()
# 读取数据
self.data = get_data()
from data.preprocess import get_data, get_data_keras
import numpy as np
from skimage.transform import resize
from operator import itemgetter
from models import log_reg
from sklearn.metrics import accuracy_score
# Validation Examples
n_valid = 1000 # Max 100000
# Model Configuration
learning_rate = 0.09
n_epochs = 1000
batch_size = 1
dataset = get_data(n_valid)
X_valid, y_valid = dataset
X_valid = X_valid * 1.0
print("Building logistic regression model using MNIST dataset...")
#log_reg.sgd_optimization_MNIST();
print("Done.")
print("Downsizing validation set...")
X_valid = resize(X_valid)
print("Done.")
print("Predicting...")
predictions = log_reg.predict_MNIST(X_d)
print("Accuracy score:")
print accuracy_score(y_valid, predictions)
#Other configuration
savename = "cnn_task3_attempt1"
###
###
print("Building model...")
model = convnets.build_cnn(input_shape, nb_classes, depth, kernel_size,
pool_size, nb_filters, dropout, activation)
print("Compiling loss function")
model.compile(loss=loss_function, optimizer=optimizer, metrics=['accuracy'])
print("Retrieving and augmenting data...")
X, Y = get_data(num_examples)
if validation:
#Splitting data in train and validation set
train_set, valid_set = utils.split_train_valid(zip(X, Y))
X_train, y_train = map(np.array, zip(*train_set))
X_valid, y_valid = map(np.array, zip(*valid_set))
X_valid = X_valid.reshape(X_valid.shape[0], num_channels, img_rows,
img_cols)
X_valid = X_valid.astype('float32')
X_valid /= 255
print(X_train.shape[0], 'train samples')
print(X_valid.shape[0], 'test samples')
# convert class vectors to binary class matrices
Y_valid = np_utils.to_categorical(y_valid, nb_classes)
else:
from data.preprocess import get_data, get_data_keras
from models import log_reg
import numpy as np
import six.moves.cPickle as pickle
from skimage.transform import resize
from sklearn.metrics import accuracy_score
# Training Configuration
n_train = 100000 # Max 100000
n_perturbed = 0
# Model Configuration
learning_rate = 0.09
n_epochs = 1000
batch_size = 1
load = False
# Get data
dataset = get_data(n_train, n_perturbed)
X, y = dataset
dataset_train = X[1000:], y[1000:]
print("Building logistic regression model")
log_reg.sgd_optimization(dataset_train, learning_rate, n_epochs, batch_size)
X_valid, y_valid = X[:1000], y[:1000]
predictions = log_reg.predict(X_valid)
print accuracy_score(y_valid, predictions)
from data.preprocess import get_data
import numpy as np
from keras.models import model_from_json
from skimage.transform import resize
from operator import itemgetter
# load json and create model
json_file = open('../trained_models/cnn1.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights("../trained_models/cnn1.h5")
print("Loaded model from disk")
X, Y = get_data()
num_example = len(X)
#Downsizing...
XX = np.zeros(shape=(num_example, 1, 28, 28), dtype="float32")
for i, x in enumerate(X):
XX[i] = resize(x, (28, 28))
outs = loaded_model.predict(XX)
def argmax2(x):
a = [(i, j) for i, j in enumerate(x)]
a = sorted(a, key=itemgetter(1))
return (a[0][0], a[1][0])