Difference between training and test


I just started learning TMVA, and noticed that in many of the guides they use the words training and test, which I could not exactly understand in the context of TMVA.
I hope someone could clear that up a bit for me.


Hey there,

When you teach a machine, you need to separate the training data and the test data.

The training data is the data used to teach your AI to make correct predictions.
The test data is the data which the machine is not allowed to be trained from, it is exclusively used for testing the goodness of the AI.

Now why is it necessary to separate them? If your machine got access to the test data, when trained long enough, it could just “memorize” the answers of the tests without making correct predictions.


I see, thank you Adam!

Also, could you maybe explain what weights are used for?


The weights are the parameters that the learning process can change to improve the prediction performance of the AI.

For example, for 1 neuron in a neural network it will have some inputs and some outputs. The inputs will be multiplied by some weights and summed and run through an activation function. Now if the weights are random, they may very well produce an output far from what we want. We can then tweak the weights a little to get closer to the output we want. We then repeat this many times for all inputs in the training set so that in the end the neuron is very good at predicting the output given the input.

Hope this clarifies a bit!