Accuracy and loss monitoring in TensorFlow - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of monitoring accuracy during model training?

Monitoring accuracy helps you see how well your model is learning to make correct predictions on the training data.

beginner

What does loss represent in model training?

Loss measures how far the model's predictions are from the true answers. Lower loss means better predictions.

intermediate

How can you monitor accuracy and loss in TensorFlow during training?

You can pass metrics=['accuracy'] to model.compile() and TensorFlow will show accuracy and loss after each training epoch.

intermediate

Why is it important to monitor both accuracy and loss?

Accuracy shows correct predictions, but loss shows how confident the model is. Monitoring both gives a fuller picture of training.

advanced

What might it mean if accuracy improves but loss does not decrease?

It could mean the model is getting more predictions right but with less confidence or uneven errors. This needs careful checking.

Which metric shows how many predictions are correct during training?

A Accuracy

B Loss

C Learning rate

D Epoch count

Accuracy measures the percentage of correct predictions.

What does a decreasing loss value during training indicate?

A Model predictions are getting closer to true values

B Model is overfitting

C Training data is increasing

D Accuracy is decreasing

Lower loss means the model's predictions are closer to the actual answers.

In TensorFlow, where do you specify to track accuracy during training?

A In model.evaluate() parameters

B In model.fit() callbacks

C In model.compile() under metrics

D In the optimizer settings

You add metrics=['accuracy'] in model.compile() to monitor accuracy.

If accuracy is high but loss is also high, what might this suggest?

A Model is confident but often wrong

B Model is correct but with low confidence

C Model is underfitting

D Training stopped early

High accuracy with high loss can mean correct predictions but low confidence or uneven errors.

Why monitor loss besides accuracy?

A Loss is only for testing

B Loss counts correct predictions

C Loss controls training speed

D Loss shows prediction confidence and error size

Loss measures how far predictions are from true values, showing confidence and error size.

Explain how accuracy and loss monitoring help during model training.

Think about what each metric tells you about the model's learning.

Describe how to set up accuracy monitoring in TensorFlow model training.

Focus on the compile step where you tell TensorFlow what to track.