def softMax(N, D, K):
# Start a session
sess = tf.InteractiveSession()
# Generate data, total no. of data = N*K; data is [N*K, D]; label is [N*K, 1]
[data, initLabel] = gen.genData(N, D, K)
# Reform label, each class is represent by a vector
label = np.zeros((N*K,K))
for i in range(N*K):
label[i][initLabel[i]] = 1
# Build the computation graph by creating nodes for the input and target output classes
x = tf.placeholder("float", shape=[None, D])
y_ = tf.placeholder("float", shape=[None, K])
# Define weights and Bias
W = tf.Variable(tf.zeros([D,K]))
b = tf.Variable(tf.zeros([K]))
# Initialize all variables
sess.run(tf.initialize_all_variables())
# Define prediction function y
y = tf.nn.softmax(tf.matmul(x,W) + b)
# Define cost function
cross_entropy = -tf.reduce_sum(y_*tf.log(y))
# Train model
train_step = tf.train.GradientDescentOptimizer(0.0001).minimize(cross_entropy)
for i in range(5000):
train_step.run(feed_dict={x: data, y_: label})
# Evaluation
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print accuracy.eval(feed_dict={x: data, y_: label})
def all_test():
train_data = ImageDataGenerator(rescale=1. / 255)
test_data = ImageDataGenerator(rescale=1. / 255)
train_generator = train_data.flow_from_directory(directory=FILE_PATH_TRAIN,
target_size=(IMG_SIZE[0],
IMG_SIZE[1]),
batch_size=BATCH_SIZE,
class_mode='categorical')
test_generator = test_data.flow_from_directory(directory=FILE_PATH_TEST,
target_size=(IMG_SIZE[0],
IMG_SIZE[1]),
batch_size=BATCH_SIZE,
class_mode='categorical')
i, j = train_generator.next()
print(i)
train = genData("D:/data/Dog-cat/train/")
i, j = train.__next__()
print(i['the_input'])
Returns:
A scalar int32 tensor with the number of examples (out of batch_size)
that were predicted correctly.
"""
# For a classifier model, we can use the in_top_k Op.
# It returns a bool tensor with shape [batch_size] that is true for
# the examples where the label's is was in the top k (here k=1)
# of all logits for that example.
correct = tf.nn.in_top_k(logits, labels, 1)
# Return the number of true entries.
return tf.reduce_sum(tf.cast(correct, tf.int32))
# number of points per class
N = 100
[data, labels] = gen.genData(N, DIMENSION, NUM_CLASSES)
[test_data, test_labels] = gen.genData(N, DIMENSION, NUM_CLASSES)
# Tell TensorFlow that the model will be built into the default Graph.
with tf.Graph().as_default():
# Generate placeholders for the images and labels.
data_placeholder = tf.placeholder(tf.float32, shape=(None, DIMENSION))
labels_placeholder = tf.placeholder(tf.int32, shape=(None))
# Build a Graph that computes predictions from the inference model.
logits = inference(data_placeholder, FLAGS.hidden1, FLAGS.hidden2)
# Add to the Graph the Ops for loss calculation.
loss = loss(logits, labels_placeholder)
# Add to the Graph the Ops that calculate and apply gradients.
train_op = training(loss, FLAGS.learning_rate)
def train(model):
"""
选择训练方式
:param model: 0,1,2。0用自己写的cnn来训练；1用MobileNetV2训练；2用ResNet34训练
:return: 训练结果的历史记录
"""
_input = Input(shape=IMG_SIZE, name='the_input')
#y_pred = cnn_layer(inputs=_input, n_class=N_CALSS)
if model == 0:
y_pred = cnn_layer(inputs=_input,
n_class=N_CALSS,
last_layer_activation=LAST_LAYER_ACTIVATION)
elif model == 1:
network = MobileNet(IMG_SIZE, N_CALSS)
y_pred = network.build(_input=_input,
last_layer_activation=LAST_LAYER_ACTIVATION)
elif model == 2:
y_pred = resnet34(inputs=_input,
n_class=N_CALSS,
last_layer_activation=LAST_LAYER_ACTIVATION)
else:
raise (TypeError("The param 'model' must in [0, 1, 2]!"))
model = Model(inputs=_input, outputs=y_pred)
#opt = SGD(lr=0.001, momentum=0.5, decay=1e-6)
#opt = Adam(lr=0.001, decay=1e-6)
#opt = RMSprop(lr=0.0001, decay=1e-6)
# 加载与训练模型
if os.path.exists(PRE_MODEL_PATH):
print("Loading model weights...")
basemodel = Model(inputs=_input, outputs=y_pred)
basemodel.summary()
basemodel = basemodel.load_weights(PRE_MODEL_PATH)
print("Done!")
# 以下是可选的loss，其计算方法请查询官方文档。也可以自定义计算loss的方法。
# binary_crossentropy适合输出标签为1个的、categorical_crossentropy、sparse_categorical_crossentropy、poisson、kl_divergence
# mean_squared_error、mean_absolute_error、mean_absolute_percentage_error、mean_squared_logarithmic_error、cosine_similarity、huber、log_cosh
# hinge、squared_hinge、categorical_hinge
model.compile(loss=sofrmaxLoss, optimizer='adam', metrics=['accuracy'])
"""
train_data = ImageDataGenerator(rescale=1. / 255)
test_data = ImageDataGenerator(rescale=1. / 255)
train_generator = train_data.flow_from_directory(
directory=FILE_PATH_TRAIN,
target_size=(IMG_SIZE[0], IMG_SIZE[1]),
batch_size=BATCH_SIZE,
class_mode='categorical')
test_generator = test_data.flow_from_directory(
directory=FILE_PATH_TEST,
target_size=(IMG_SIZE[0], IMG_SIZE[1]),
batch_size=BATCH_SIZE,
class_mode='categorical')
"""
train_generator = genData("D:/data/Dog-cat/train")
test_generator = genData("D:/data/Dog-cat/test")
save_dir = SAVE_DIR
if not os.path.exists(save_dir):
print("Making Dir:", save_dir)
os.mkdir(save_dir)
checkpoint = ModelCheckpoint(filepath=save_dir +
'/test-{epoch:02d}-{val_loss:.2f}.h5',
monitor='val_loss',
save_best_only=False,
save_weights_only=True)
lr_schedule = lambda epoch: 0.0005 * 0.4**epoch
learning_rate = np.array([lr_schedule(i) for i in range(EPOCH)])
changelr = LearningRateScheduler(lambda epoch: float(learning_rate[epoch]))
tensorboard = TensorBoard(log_dir=save_dir + '/logs', write_graph=True)
print("---------------start training---------------")
history = model.fit_generator(
train_generator,
steps_per_epoch=24000 // BATCH_SIZE,
epochs=EPOCH,
initial_epoch=0,
validation_data=test_generator,
validation_steps=1000 // BATCH_SIZE,
callbacks=[checkpoint, changelr, tensorboard])
return history.history
#output = Model(inputs=_input, outputs=model.get_layer(name='conv1_1').output)
#print(output.predict(_input))
"""
