方法一：

keras.utils.vis_utils模組提供了畫出Keras模型的函式（利用graphviz）
然而模型視覺化過程會報錯誤：

from keras.utils import plot_model
plot_model(model, to_file='model.png')

keras文件給出的解決方法：

pip install pydot-ng & brew install graphviz

安裝時會提醒你新增環境變數：

You may want to update following environments after installed linuxbrew.

  PATH, MANPATH, INFOPATH 
 

開啟.bashrc：

gedit ~/.bashrc

在最後新增提示的環境變數即可
如果已經安裝.linuxbrew ，若提示錯誤，可以把.linuxbrew刪除再繼續安裝
詳細homebrew在Linux下的使用討論及Linuxbrew安裝方法

方法二 :

開啟keras視覺化程式碼：

def _check_pydot():
    try:
        # Attempt to create an image of a blank graph
        # to check the pydot/graphviz installation.
        pydot.Dot.create(pydot.Dot())
    except
 
 Exception:
        # pydot raises a generic Exception here,
        # so no specific class can be caught.
        raise ImportError('Failed to import pydot. You must install pydot'
                          ' and graphviz for `pydotprint` to work.')

可自行pip安裝：

sudo apt-get install graphviz
sudo pip install pydot-ng
 

注意需要先安裝graphviz再裝pydot-ng

視覺化結果

隨便寫了一個2層LSTM的網路：

from keras.models import Model
from keras.layers import LSTM, Activation, Input
import numpy as np
from keras.utils.vis_utils import plot_model

data_dim = 1
timesteps = 12
num_classes = 4

inputs = Input(shape=(12,1))
lstm1 = LSTM(32, return_sequences=True)(inputs)
lstm2 = LSTM(4 , return_sequences=True)(lstm1)
outputs = Activation('softmax')(lstm2)
model = Model(inputs=inputs,outputs=outputs)
model.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

x_train = np.random.random((1000, timesteps, data_dim))
y_train = np.random.random((1000, timesteps, num_classes))

x_val = np.random.random((100, timesteps, data_dim))
y_val = np.random.random((100, timesteps, num_classes))

model.fit(x_train, y_train,
          batch_size=64, epochs=5,
          validation_data=(x_val, y_val))
#模型視覺化
plot_model(model, to_file='model.png')
x = np.arange(12).reshape(1,12,1)
a = model.predict(x,batch_size=64)
print a

結果：