The Flipbook is our user interface for viewing your model training in real time. In the flipbook you can watch the model progress through epochs and view metrics along the way. Each epoch is the model taking in more training data and making connections using deep learning neural networks.
To put it more simply: If you are training a model to identify a dog, the model will take a batch of images and start to learn what a dog is. In this first batch it might recognize that dogs have 4 legs and short hair. In the second batch of imagery (next epoch) it might see training data of dogs with long hair but also four legs, and the model will start to make connections from the first set to the second set that it’s a dog because there are 4 legs but now also dogs can have short or long fur. In the next batch of images (3rd epoch) the model might see data of dogs with different color coats and make connections to the previous epochs that these new coat colors have both long and short hair examples and now understand that dogs can come in a variety of coat patterns, but it is still a dog.
This is the type of model development you can see in real time via the flipbook!
During the training in the flipbook, you can see datapoints moving corresponding to the training and validation data the model is using to train on. When you click train on the platform the data you have labeled (training data) is automatically split into 80/20 training and validation data. The validation data, which is labeled, will act as ground truth for scoring the model to create metrics. You can click on each datapoint to see even more insight to the model training. This is important for understanding outliers in your data— understanding your outliers in your data will help you to create a more diverse dataset, in turn creating a more robust model.
The flipbook also exposes model metrics:
Precision measures the accuracy of the positive predictions. It is the ratio of correctly predicted positive observations to the total predicted positives. High precision indicates that when the model predicts positive, it is often correct. This is important in cases where the cost of a false positive is high.
Recall is known as the sensitivity or true positive rate, as it measures the model’s ability to capture all the actual positive cases. It is the ratio of correctly predicted positive observations to all observations in the actual class. High recall indicates that the model can identify most of the positive cases. It is crucial when missing a positive case (a false negative) has a high cost.
F1 Score is the harmonic mean of precision and recall. It balances the two metrics and is especially useful when the class distribution is imbalanced, i.e., one class is more frequent than the other. The F1 Score provides a single metric that considers both precision and recall, offering a balance between the two. It is particularly useful when you need to strike a balance between precision and recall.
Here is an overview of the training flip book and all of its key functions!
At the bottom of the flipbook, you can see all of the model metrics which pretrain to your model
Here are some examples of what that might look like:
