Quiz - 23/24 Flashcards
1
Q
If the accuracy is on the training set is high, but the accuracy on the test set is low, which of the following strategies are reasonable?
A
- Use a larger training set
- Increase regularisation
- Increase dropout
2
Q
What datasets are suited for kernel methods?
A
Datasets with small n and large p
3
Q
Which of these is the best tool to decide which threshold to use on a Logistic Regression binary classifier?
A
ROC
4
Q
Which of the following statements are correct?
- GD is guaranteed to converge to global or local min.
- Mini-batch size has no influence on convergence speed.
- Learning rate is high, conv. will be faster.
- Learning rate high, GD might overshoot.
A
3, 4
5
Q
What does a high value of logit mean in logistic regression
A
- Observation is far from decision boundary.
- Likelihood for the observation to belong to class 1 is high.
6
Q
What is a unit?
A
A unit is a differentiable function.
7
Q
What is an artificial neuron?
A
A linear combination of i/p followed by an activation function.
8
Q
When is a neural network ‘deep’
A
A neural network is deep if it is made of, at least, 2 hidden layers with non-linear activations.
9
Q
A
