# How to evaluate te testing data from the trained data?

I am new at ML, and still trying to understand some concepts, so I figured I could ask here and maybe finally understand.

How does it work the whole splitting data from a data set?

Before answering, here is what I "know" and "understand" so far:

I have a data set, I'll call it "ds". ds has 10 features, f1,f2...f10, would be represented by each column of the ds. Now, I also have an array of labels that goes with each row of ds, which represents what each sample of data outputs.

Samples F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 Labels
1 ele1 ele2 ele3 ele4 ele5 ele6 ele7 ele8 ele9 ele10 1
2 ele11 ele12 ele13 ele14 ele15 ele16 ele17 ele18 ele19 ele20 0
... ... ... ... ... ... ... ... ... ... ... ...
n n n+1 n+2 n+3 n+4 n+5 n+6 n+7 n+8 n+9 1

Next, when testing ML algorithms, I'm suppose to split the ds into training data (training_data) and testing data (testing_data), and same thing with the labels, training label (training_labels), and testing labels (testing_labels). So now I have 4 pieces of information.

In order to check the algorithm's performance, I have to run the algorithm to each set of training data (rows...), and it would produce one result for each row, to which I would like to call "trained data". Once the algorithm has been run on the training data, I will also have an array of "trained_labels" ->[trained_label1, trained_label2, ... trained_label_n ] which it's size will match with the number of rows of the training_data.

Finally, I'm supposed to use this newly created "trained_labels" and compare it to the testing_labels and based on the accuracy (if all the trained_labels match all the testing_labels then it would be 100% accurate), I will be able to tell how good was the algorithm

Assuming a proportion 80% training and 20% testing, here is the big question that I still I am not able to wrap my head around, the size of the trained_labels will be different to the size of the testing labels, so given that fact, how can I campare the accuracy?

Like coding that with python syntax would be:

accuracy = (testing_labels == trained_labels).sum() / len(testing_labels) *100,

So if I could get some guidance it would be very appreciated.

Kind regards

P.S: I don't think slicing the training_labels size to match testing_labels size would be the answer.