python - 使用 Hualos 可视化 Keras 中的训练进度

Question

我正在尝试按照此处的说明https://github.com/fchollet/hualos 来可视化 Keras 中的训练进度。在上述页面中，我读到：

Hualos - Keras 全面可视化项目

目前，这是一个简单的演示，其中 Flask 服务器公开了一个 API，以 JSON 对象的形式发布和使用事件。Keras 回调 RemoteMonitor 可以向服务器发布事件，Hualos 登陆页面监听服务器并在 c3.js 图表上显示传入数据。

例子：

start the server: python api.py
load the landing page: http://localhost:9000/
launch a Keras experiment with the RemoteMonitor callback:

> from keras import callbacks
> remote = callbacks.RemoteMonitor(root='http://localhost:9000')
> 
> model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
> validation_data=(X_test, Y_test), callbacks=[remote])

而且：

依赖项：

Python:
    Flask
    gevent

JS (included in the repo):
    d3.js
    c3.js

我已经成功安装了 Flask 和 gevent。

为简洁起见，我的代码如下使用 mnist 测试数据，我从这里下载为 csv 文件：https ://pjreddie.com/projects/mnist-in-csv/

from keras import callbacks
remote = callbacks.RemoteMonitor(root='http://localhost:9000')

X = mnist.iloc[:, 1:].values
y = to_categorical(mnist.iloc[:, 0])

X =  X.astype('float32')
y =  y.astype('float32')

X /= 255  # ATTENTION!  Normalization is critical!!!

n_cols = X.shape[1]

# Create the model: model
model = Sequential()

# Add the first hidden layer
model.add(Dense(50, activation = 'relu', input_shape = (784,)))

# Add the second hidden layer
model.add(Dense(50, activation = 'relu'))

# Add the output layer
model.add(Dense(10, activation = 'softmax'))

# Compile the model
model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])

# Fit the model
model.fit(X, y, validation_split = 0.3, epochs=3, callbacks = [remote])

这导致以下输出：

Train on 7000 samples, validate on 3000 samples
Epoch 1/3
7000/7000 [==============================] - 1s 100us/step - loss: 0.7812 - acc: 0.7717 - val_loss: 0.3036 - val_acc: 0.9153
Epoch 2/3
3296/7000 [=============>................] - ETA: 0s - loss: 0.3139 - acc: 0.9072

C:\ProgramData\Anaconda3\lib\site-packages\keras\callbacks.py:606: UserWarning: Warning: could not reach RemoteMonitor root server at http://localhost:9000
  'root server at ' + str(self.root))

7000/7000 [==============================] - 0s 55us/step - loss: 0.3051 - acc: 0.9111 - val_loss: 0.2616 - val_acc: 0.9213
Epoch 3/3
2400/7000 [=========>....................] - ETA: 0s - loss: 0.2613 - acc: 0.9237

C:\ProgramData\Anaconda3\lib\site-packages\keras\callbacks.py:606: UserWarning: Warning: could not reach RemoteMonitor root server at http://localhost:9000
  'root server at ' + str(self.root))

7000/7000 [==============================] - 0s 63us/step - loss: 0.2397 - acc: 0.9284 - val_loss: 0.2350 - val_acc: 0.9320

C:\ProgramData\Anaconda3\lib\site-packages\keras\callbacks.py:606: UserWarning: Warning: could not reach RemoteMonitor root server at http://localhost:9000
  'root server at ' + str(self.root))

事实上，当我尝试运行命令时：

python api.py

我遇到了一个例外：

 (base) D:\Mint_ns>python api.py
Traceback (most recent call last):
  File "api.py", line 10, in <module>
    from pattern.server import App
ModuleNotFoundError: No module named 'pattern'

(base) D:\Mint_ns>conda install -c asmeurer pattern
Solving environment: failed

UnsatisfiableError: The following specifications were found to be in conflict:
  - pattern
  - tensorflow
Use "conda info <package>" to see the dependencies for each package.

这有点奇怪，因为重点是使用 Hualos 与以 TensorFlow 作为后端运行的 Keras 一起工作。

我怎样才能使这项工作？

标签： pythonkerasvisualizationdiagnostics