← Back to Blog

Understanding the Use of Verbose in Keras Model Validation

In this blog, we will learn about the abundant features provided by Keras, a widely-used deep learning library, designed to simplify the workflow for data scientists. Specifically, we will explore the functionality of the verbose argument within Keras' model training and validation methods. This post aims to provide insights into leveraging the verbose parameter for effective model validation in Keras.
By Saturn Cloud | | Miscellaneous

Keras, a popular deep learning library, offers a plethora of features to make the life of a data scientist easier. One such feature is the verbose argument in model training and validation methods. This blog post will delve into the use of verbose in Keras while validating the model.

Table of Contents

  1. What is Verbose in Keras?
  2. Why Use Verbose in Model Validation?
  3. How to Use Verbose in Keras?
  4. Common Errors and How to Handle Them
  5. Conclusion

What is Verbose in Keras?

Verbose Levels

In Keras, verbose is an optional argument in various methods, including fit(), evaluate(), and predict(). It is used to set the logging level during the model training and validation process. The verbose argument can take one of three integer values: 0, 1, or 2.

  • verbose=0: Silent mode - no output during training.
  • verbose=1: Progress bar mode - displays a progress bar with training and validation metrics (default).
  • verbose=2: One line per epoch - shows a summary of training and validation metrics after each epoch.

Choosing the Right Verbose Level

Selecting the appropriate verbose level depends on your preference and the level of detail you need. For quick insights, you might prefer verbose=1, while for a more detailed overview, verbose=2 can be useful.

Why Use Verbose in Model Validation?

Verbose plays a crucial role in model validation for several reasons:

  1. Monitoring Progress: Verbose provides real-time feedback about the model’s performance during validation. This can be particularly useful when validating large models or datasets that can take a significant amount of time to process.

  2. Debugging: If your model isn’t performing as expected, verbose logs can provide valuable insights into what might be going wrong. For example, if your model’s loss isn’t decreasing as expected, verbose logs can help identify if the issue is due to overfitting, underfitting, or a problem with the data itself.

  3. Performance Analysis: Verbose logs provide detailed information about each epoch, including the loss and any metrics you’ve chosen to monitor. This can be useful for analyzing your model’s performance and making decisions about how to improve it.

How to Use Verbose in Keras?

Using verbose in Keras is straightforward. When calling the fit(), evaluate(), or predict() methods, you simply pass in the desired verbose level as an argument. Here’s an example:

# Import Keras and datasets
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense

# Load data
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

# Preprocess data
train_images = train_images.reshape((60000, 28 * 28))
train_images = train_images.astype('float32') / 255

test_images = test_images.reshape((10000, 28 * 28))
test_images = test_images.astype('float32') / 255

# Define model
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(28 * 28,)))
model.add(Dense(10, activation='softmax'))

# Compile model
model.compile(optimizer='rmsprop',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

# Fit model with verbose=1
model.fit(train_images, train_labels, epochs=5, batch_size=128, verbose=1)

# Evaluate model with verbose=0
test_loss, test_acc = model.evaluate(test_images, test_labels, verbose=0)

print('Test accuracy:', test_acc)

In this example, model.fit() is called with verbose=1, which means a progress bar with logs will be displayed during training. Then, model.evaluate() is called with verbose=0, so no logs will be displayed during validation.

Here’s the expected output for the training part (model.fit) with verbose=1:

Epoch 1/5
469/469 [==============================] - 2s 3ms/step - loss: 0.2565 - accuracy: 0.9269
Epoch 2/5
469/469 [==============================] - 2s 3ms/step - loss: 0.1044 - accuracy: 0.9687
Epoch 3/5
469/469 [==============================] - 2s 3ms/step - loss: 0.0694 - accuracy: 0.9792
Epoch 4/5
469/469 [==============================] - 2s 3ms/step - loss: 0.0508 - accuracy: 0.9850
Epoch 5/5
469/469 [==============================] - 2s 3ms/step - loss: 0.0385 - accuracy: 0.9889

For the evaluation part (model.evaluate) with verbose=0, there won’t be any output during the evaluation. The result will only be stored in the test_loss and test_acc variables, which are then printed:

Test accuracy: 0.9779

Common Errors and How to Handle Them

Error 1: Unresponsive Output

If the output becomes unresponsive, it may be due to high verbosity levels. Consider reducing the verbosity or using a less detailed logging mechanism.

Error 2: Inconsistent Progress Updates

Inconsistencies in progress updates may occur with custom callbacks. Ensure that your callback implementation is correct and compatible with the chosen verbosity level.

Conclusion

The verbose argument in Keras is a powerful tool for monitoring and debugging your models during the validation process. By understanding and effectively using this feature, data scientists can gain deeper insights into their models' performance and make more informed decisions about how to improve them.

About Saturn Cloud

Saturn Cloud is your all-in-one solution for data science & ML development, deployment, and data pipelines in the cloud. Spin up a notebook with 4TB of RAM, add a GPU, connect to a distributed cluster of workers, and more. Request a demo today to learn more.

Get a Technical Demo