绘制keras自定义模型时，"ListWrapper"对象没有属性“名称" [英] 'ListWrapper' object has no attribute 'name' when plotting keras custom model

问题描述

我想绘制一个定制keras模型的基础拓扑.根据此链接( https://machinelearningmastery.com/visualize-我认为我可以只使用keras.utils.vis_utils.plot_model，但这会产生错误.

I would like to plot the underlying topology of a custom keras model. According to this link (https://machinelearningmastery.com/visualize-deep-learning-neural-network-model-keras/) I thought I'd be able to just use keras.utils.vis_utils.plot_model, but this yielded an error.

这是最小的自定义模型和代码来重现该错误:

Here's a minimal custom model and code to reproduce the error:

import tensorflow as tf

from keras.models import Model
from keras import backend as K
from keras.utils.vis_utils import plot_model

import unittest

'''
Construct a double-layer perceptron without an activation
'''

rows = 10
cols = 2

class Model(tf.keras.Model):
    def __init__(self, hidden_topology):
        super(Model, self).__init__(name='')
        self.hidden_topology = hidden_topology

    def call(self, inputs):
        hidden_output = inputs
        for hidden_layer in self.hidden_topology:
            hidden_output = hidden_layer(hidden_output)

        return hidden_output

    def compute_output_shape(self, input_shape):
        return (input_shape[0][0], 1)

model = Model(
    [
        tf.keras.layers.Dense(
            1,
            input_shape=((rows, cols), ),
            use_bias=True,
            kernel_initializer=tf.constant_initializer(1.0),
            bias_initializer=tf.constant_initializer(0.0)),
        tf.keras.layers.Dense(
            1,
            input_shape=((rows, cols), ),
            use_bias=True,
            kernel_initializer=tf.constant_initializer(1.0),
            bias_initializer=tf.constant_initializer(0.0))
    ])


test_data = np.reshape(range(rows*cols), (rows,cols)).astype(np.float32)
top = model.call(test_data)

#plot_model(top, to_file='model_plot.png')#, show_shapes=True, show_layer_names=True)
plot_model(model, to_file='model_plot.png')#, show_shapes=True, show_layer_names=True)

这会产生以下错误:

AttributeErrorTraceback (most recent call last)
<ipython-input-3-b73c347c7b0a> in <module>()
     49 # top = model.call(test_data)
     50 
---> 51 plot_model(model, to_file='model_plot.png')#, show_shapes=True, show_layer_names=True)
     52 
     53 # def call(self, inputs):

/package/python-2.7.15/lib/python2.7/site-packages/keras/utils/vis_utils.pyc in plot_model(model, to_file, show_shapes, show_layer_names, rankdir, expand_nested, dpi)
    238     """
    239     dot = model_to_dot(model, show_shapes, show_layer_names, rankdir,
--> 240                        expand_nested, dpi)
    241     _, extension = os.path.splitext(to_file)
    242     if not extension:

/package/python-2.7.15/lib/python2.7/site-packages/keras/utils/vis_utils.pyc in model_to_dot(model, show_shapes, show_layer_names, rankdir, expand_nested, dpi, subgraph)
    104 
    105         # Append a wrapped layer's label to node's label, if it exists.
--> 106         layer_name = layer.name
    107         class_name = layer.__class__.__name__
    108 

AttributeError: 'ListWrapper' object has no attribute 'name'

我也尝试了注释行，但无济于事.

I also tried the commented out line, to no avail.

如何可视化此拓扑?我正在使用tensorflow 2.0.0

How can I visualise this topology? I'm using tensorflow 2.0.0

推荐答案

在使用tf.keras时，您提到的链接正在使用keras(Tensorflow的高级

Link you have mentioned is using keras while you are using tf.keras(Tensorflow's high level API).
Instead of:

from keras.utils.vis_utils import plot_model

将此行更改为:

from tensorflow.keras.utils import plot_model

修改:
尽管您将摆脱此错误，但是由于您使用的是子类化模型，因此您只能在绘图中看到一个模型块.要绘制完整的模型图，您将必须使用 Sequential 或功能模型.我还建议将班级名称更改为Model以外的其他名称.

Although you will get rid of this error, but since you are using sub-classed model all you will see is a model block in your plot.To plot complete model graph you'll have to use either Sequential or Functional model. I would also suggest to change name of your class to something other than Model.

这篇关于绘制keras自定义模型时，"ListWrapper"对象没有属性“名称"的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！