Keras 模型编译¶

model.compile(loss,
              optimizer,
              metrics)

1. Losses¶

Loss Function

Objective Function

Optimization score Function

(1) Keras loss functions

回归:

from keras import losses

# 回归
from keras.losses import mean_squared_error
from keras.losses import mean_absolute_error
from keras.losses import mean_absolute_percentage_error
from keras.losses import mean_squared_logarithmic_error
from keras.losses import squared_hinge
from keras.losses import hinge
from keras.losses import categorical_hinge
from keras.losses import logcosh

model.Compile(loss = ["mse", "MSE", mean_squared_error],
                       optimizer,
                       metircs)
model.Compile(loss = ["mae", "MAE", mean_absolute_error],
                       optimizer,
                       metircs)
model.Compile(loss = ["mape", "MAPE", mean_absolute_percentage_error],
                       optimizer,
                       metircs)
model.Compile(loss = ["msle", "MLSE", mean_squared_logarithmic_error],
                       optimizer,
                       metircs)

分类:

# 分类
from keras.losses import categorical_crossentropy
from keras.losses import sparse_categorical_crossentropy
from keras.losses import binary_crossentropy
from keras.losses import kullback_leibler_divergence
from keras.losses import poisson
from keras.losses import cosine_proximity

model.Compile(loss = ["kld", "KLD", kullback_leibler_divergence],
                       optimizer,
                       metircs)
model.Compile(loss = ["cosine", cosine_proximity],
                       optimizer,
                       metircs)

2. Metrics¶

Metric 是一个评估模型表现的函数

Metric 函数类似于一个损失函数，只不过模型评估返回的 metric 不用来训练模型，因此，可以使用任何损失函数当做一个 metric 函数使用

2.1 Keras metrics¶

Keras API:

from keras import metrics
from keras.metrics import binary_accuracy
from keras.metrics import categorical_accuracy
from keras.metrics import sparse_categorical_accuracy
from keras.metrics import top_k_categorical_accuracy
from keras.metrics import sparse_top_k_categorical_accuracy
from keras.metrics import mae

from keras.losses import mean_squared_error
from keras.losses import mean_absolute_error
from keras.losses import mean_absolute_percentage_error
from keras.losses import mean_squared_logarithmic_error
from keras.losses import squared_hinge
from keras.losses import hinge
from keras.losses import categorical_hinge
from keras.losses import logcosh
from keras.losses import categorical_crossentropy
from keras.losses import sparse_categorical_crossentropy
from keras.losses import binary_crossentropy
from keras.losses import kullback_leibler_divergence
from keras.losses import poisson
from keras.losses import cosine_proximity

Metrics Name:

metrics = ["acc", "accuracy"]

2.3 自定义 metrics¶

import keras.backend as K

def mean_pred(y_true, y_pred):
     return K.mean(y_pred)

model.compile(optimizers = "rmsprop",
                       loss = "binary_accuracy",
                       metrics = ["accuracy", mean_pred])

3. Optimizers¶

3.1 Keras optimizder 的使用方式¶

keras.optimizers 和 optimizer 参数

from keras import optimizers

# 编译模型
sgd = optimizers.SGD(lr = 0.01, decay = 1e-6, momentum = 0.9, nesterov = True)
model.compile(loss = "mean_squared_error",
              optimizer = sgd)

optimizer 参数

# 编译模型
model.compile(loss = "mean_squared_error",
              optimizer = "sgd")

3.2 Keras optimizers 的共有参数¶

control gradient clipping
- clipnorm
- clipvalue

from keras import optimizers

# All parameter gradients will be clipped to
# a maximum norm of 1.
sgd = optimizers.SGD(lr = 0.01, clipnorm = 1)

# All parameter gradients will be clipped to
# a maximum value of 0.5 and
# a minimum value of -0.5.
sgd = optimizers.SGD(lr = 0.01, clipvalue = 0.5)

3.3 Keras Optimizers¶

SGD
RMSprop
Adagrad
Adadelta
Adam
Adamax
Nadam

from keras import optimizers

sgd = optimizers.SGD(lr = 0.01)
model.compile(loss, optimizer = sgd)
# or
model.compile(loss, optimizer = "sgd")

rmsprop = optimizers.RMSprop(lr = 0.001)
model.compile(loss, optimizer = rmsprop)
# or
model.compile(loss, optimizer = "rmsprop")

adagrad = optimizers.Adagrad(lr = 0.01)
model.compile(loss, optimizer = adagrad)
# or
model.compile(loss, optimizer = "adagrad")

adadelta = optimizers.Adadelta(lr = 1.0)
model.compile(loss, optimizer = adadelta)
# or
model.compile(loss, optimizer = "adadelta")

adam = optimizers.Adam(lr = 0.001)
model.compile(loss, optimizer = adam)
# or
model.compile(loss, optimizer = "adam")

adamax = optimizers.Adamax(lr = 0.02)
model.compile(loss, optimizer = adamax)
# or
model.compile(loss, optimizer = "adamax")

nadam = optimizers.Nadam(lr = 0.002)
model.compile(loss, optimizer = nadam)
# or
model.compile(loss, optimizer = "nadam")