Bob (look up homework 1 if you don't remember who he is) now has another dataset...

 

Transcribed Image Text:

Bob (look up homework 1 if you don't remember who he is) now has another dataset and wants to split it for training/testing again. This time, his dataset also contains 1,000 examples, but each input only has 1 features. This is also a binary classification dataset that is perfectly balanced (each class contains 500 examples). Although he agrees with your previous feedback (in homework 1), he still thinks that he doesn't need to split the data randomly, and thus came up with the following method to split the data: 1. First, he separated all examples from class 0 and class 1 into two disjoint subsets. Then he sorted the inputs in each of these subsets. The results of this step are 2 variables X and Z, where: X contains the sorted inputs of examples from class 0; that is, X[0] < X[1] S... ≤ X[499]. • Z contains the sorted inputs of examples from class 1; that is, Z[0] < Z[1] S... <Z(499]. 2. Then he selected the train and test sets as followed (each pair is an input-label pair): . Train set {(x[0], 0), (X[1], 0),..., (X[349], 0), (Z[0], 1), (2[1], 1),. (Z[349], 1)). = Y Test set {(X[350], 0), (X[351], 0),.. (X[499], 0), (Z[350], 1), (Z[351], 1),..., (Z[499], 1)). = Is this a good split? If not, explain what may go wrong with a specific example. Bob (look up homework 1 if you don't remember who he is) now has another dataset and wants to split it for training/testing again. This time, his dataset also contains 1,000 examples, but each input only has 1 features. This is also a binary classification dataset that is perfectly balanced (each class contains 500 examples). Although he agrees with your previous feedback (in homework 1), he still thinks that he doesn't need to split the data randomly, and thus came up with the following method to split the data: 1. First, he separated all examples from class 0 and class 1 into two disjoint subsets. Then he sorted the inputs in each of these subsets. The results of this step are 2 variables X and Z, where: X contains the sorted inputs of examples from class 0; that is, X[0] < X[1] S... ≤ X[499]. • Z contains the sorted inputs of examples from class 1; that is, Z[0] < Z[1] S... <Z(499]. 2. Then he selected the train and test sets as followed (each pair is an input-label pair): . Train set {(x[0], 0), (X[1], 0),..., (X[349], 0), (Z[0], 1), (2[1], 1),. (Z[349], 1)). = Y Test set {(X[350], 0), (X[351], 0),.. (X[499], 0), (Z[350], 1), (Z[351], 1),..., (Z[499], 1)). = Is this a good split? If not, explain what may go wrong with a specific example.