STYLE ON THE STREET (Learning)

Training a model

Learning Objective: Train and evaluate a classification or prediction model using machine learning on a tabular dataset.

Enduring Understanding: Computers can learn to classify instances or predict values by examining feature values. If the results on new inputs are unsatisfactory, additional training may be required to improve the accuracy.

Unpacked: Within a tabular dataset, each training example is a row in the table and is described by a set of feature values; the features are the columns of the table. Classification assigns each example to one of a discrete set of classes (e.g., cat or dog); prediction outputs a continuous value, such as predicting a person's height from their age. The learning algorithm is likely to be a decision tree learner rather than a neural network.

Activity: Sites like MachineLearningForKids and eCraft2Learn include decision tree learning. The learning algorithm figures out which are the relevant features and what values they should have for each class.

In the previous blog, we talked about the difference between neural networks and decision trees. A neural network is a black box, but a decision tree is more interpretable. So, it is perfect for beginners to learn about data design. Now, in this blog, I expect everyone to make their own image/text classifiers based on decision trees. It is simple and easy!

There are four types of dataset diversity strategies: perspective, contexts, and states relate to diversity on a single object, while types involved variation on a class of objects.

• Perspectives Contexts: Varying viewpoints on a single object

  • Rotated object

  • Crop and size of object

• States:Varying the background around an object

  • Color or pattern of background

  • Presence of occluding objects (e.g., hand holding item)

• Types: Varying the form factor or condition of an object

  • Items opened or closed

  • Ingredients slice or whole

  • Condition of an object (e.g., health or sick plants)

• Deinition: Varying the types of objects within a class

  • Distinct form factors (e.g., over-the- ear-headphones versus in-ear ear- buds)

Activity: Identifying fashion labels from clothing tags and providing information about the brand’s environmental impact (CHNGE, F21, H&M, Patagonia, Zara)

Let open https://machinelearningforkids.co.uk

If you have any question, check the video above!