如何可视化含输入输出层数值的DNN/CNN模型？现有可视化仅显示Dense层

Hey there! Let's work through your two technical requests one by one.

1. How to Visualize DNN/CNN Models with Input & Output Layer Values #

To visualize your model with clear display of input/output layer details (like shape, feature counts, and output dimensions), you'll want to leverage the built-in tf.keras.utils.plot_model function with specific parameters that enable these details. Here's what you need to know:

• Key Parameters to Enable Layer Details:
  • show_shapes=True: This will display the input and output tensor shapes for every layer (including your input and output layers, showing values like (None, n_features) for the input and (None, n_outputs) for the output).
  • show_layer_names=True: Makes layer names visible, which helps map each component in the plot to your code.
  • show_layer_activations=True: Adds the activation function name next to each Dense/CNN layer for clarity.
  • expand_nested=False: Keeps the plot clean if you don't have nested sub-models (adjust to True if you do).

For CNN models, the approach is identical—just make sure your Input layer specifies the full spatial shape (e.g., Input(shape=(28,28,1)) for MNIST images), and the show_shapes parameter will display those dimensions in the visualization.

2. Fixing Missing Input/Output Layer Values in Your Current Visualization #

The reason your generated image only shows Dense layers (and no input/output layer values) is that you're using the default plot_model settings, which don't display shapes by default. Here's how to modify your code to fix this:

Modified Code Snippet #

import pandas as pd
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Input, BatchNormalization, Dense, Dropout

def create_model(n_features, n_outputs):
    model = Sequential()
    model.add(Input(n_features))
    model.add(BatchNormalization())
    model.add(Dense(51, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(68, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(85, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(85, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(68, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(51, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(n_outputs, activation='sigmoid'))
    model.compile(loss='binary_crossentropy', optimizer='Adam', metrics=['accuracy'])
    
    # Updated plot_model call with shape/details enabled
    tf.keras.utils.plot_model(
        model,
        to_file='model_combined.png',
        show_shapes=True,
        show_layer_names=True,
        show_layer_activations=True
    )
    # model.summary()
    return model

Why This Works: #

• Adding show_shapes=True forces the plot to display the input/output dimensions for every layer. You'll now see your input layer's (None, n_features) shape and output layer's (None, n_outputs) shape clearly in the image.
• The other parameters (show_layer_names, show_layer_activations) add extra context to make the visualization more informative.

Note on the Earlier keras.utils.vis_utils.plot_model Error #

The TypeError: 'InputLayer' object is not iterable you encountered is likely due to version mismatches between standalone Keras and TensorFlow's Keras implementation. Since you're building your model with tf.keras.Sequential, sticking with tf.keras.utils.plot_model is the correct and compatible choice (as you've already done).

内容的提问来源于stack exchange，提问作者bashar