Get Multi-Label Model Learning Curve

Response Body

Name	Type	Description	Available Version
`data`	array	Array of `learningcurve` objects.	2.0
`object`	string	Object returned; in this case, `list`.	2.0

learningCurve Response Body

Name	Type	Description	Available Version
`epoch`	int	Epoch to which the metrics correspond.	2.0
`epochResults`	object	Prediction results for the set of data used to test the model during training.	2.0
`metricsData`	object	Model metrics values.	2.0
`object`	string	Object returned; in this case, `learningcurve`.	2.0

metricsData Response Body

Name	Type	Description	Available Version
`confusionMatrices`	object	This object contains: A string for each label in the model. For each label, a two-element array. For each element in the array, a two-element integer array that contains the confusion matrix values. These values are the correct and incorrect classifications for the label based on testing done during the training process. Use this field to build a binary confusion matrix for each label in the model.	2.0
`f1`	array	Array of float arrays that contains the weighted average of precision and recall for each label in the dataset. The corresponding label for each value in this array can be found in the `labels` array. For example, the f1 score in the first array corresponds to the first label in the `labels` array.	2.0
`labels`	array	Array of strings that contains the dataset labels. These labels correspond to the values in the `f1` array and the `confusionMatrix` array.	2.0
`testAccuracies`	array	Array of floats that specify the accuracy of the test data for each label. From your initial dataset, by default, 10% of the data is set aside and isn't used during training to create the model. This 10% is then sent to the model for prediction. How often the correct prediction is made with this 10% is reported for each label in the `testAccuracies` array.	2.0
`trainingAccuracies`	array	Array of floats that specify the accuracy of the training data for each label. By default, 90% of the data from your dataset is left after the test accuracy set is set aside. This 90% is then sent to the model for prediction. How often the correct prediction is made with this 90% is reported for each label in the `trainingAccuracies` array.	2.0

epochResults Response Body

Name	Type	Description	Available Version
`exampleName`	string	Example name, followed by a hyphen, and the expected label. For example, `"exampleName": "006_0034.jpg-006.basketball"`.	2.0
`expectedLabel`	string	Image label provided when the dataset (used to create the model) was created.	2.0
`predictedLabel`	string	Label that the model predicted for the example	2.0

You can use the learning curve data to see how a model performed in each training iteration. This information helps you tune your model and identify the optimal number of epochs to specify when you train a dataset.

When you train a dataset, you can specify the number of epochs used to create the model, or if you don't specify any epochs, the training call selects a number of epochs based on the dataset data. In each epoch, machine learning happens behind the scenes to create the model. Information from an epoch is passed into the next epoch, from that epoch to the next, and so on.

When it comes to epochs, more isn't always better. For example, you could train a dataset and specify five epochs. Using the learning curve data for the resulting model, you might see that the most accurate results for that dataset are in the third epoch. You can use the learning curve data to see if your model is overfit (model predicts accurately with training data but not unseen data) or underfit (model doesn't predict accurately with training data or unseen data).

For information about correlating the confusion matrix values and the labels, see Get Multi-Label Model Metrics.

Use the Epoch Results to Tune the Model

Use the epochResults values to better understand the testAccuracy number the model returns for each epoch. For example, if testAccuracy is 1, then each example in the set of test examples was predicted correctly. But if testAccuracy is .5, then you can look at the epochResults to see for which test data the model returned incorrect predictions.

Query Parameters

By default, this call returns 25 epochs. To page through the epochs for a model, use the offset and count query parameters.

Name	Description	Available Version
`count`	Number of epochs to return. Maximum valid value is 25. If you specify a number greater than 25, the call returns 25 epochs. Optional.	1.0
`offset`	Index of the epoch from which you want to start paging. Optional.	1.0

Here's an example of these query parameters. If you omit the count parameter, the API returns 25 epochs. If you omit the offset parameter, paging starts at 0.

curl -X GET -H "Authorization: Bearer <TOKEN>" -H "Cache-Control: no-cache"  "https://api.einstein.ai/v2/vision/models/7JPFL6UKMHUEJLHJ6FEX4YBV4U/lc?offset=30&count=10"

For example, let's say you want to page through all of the model epochs and show 10 at a time. The first call would have offset=0 and count=10, the second call would have offset=10 and count=10, and so on.