Intro to ML: The ML Revision Quiz

If we predict every observation to be True, what will our model precision be?

• 100%

• 0%

• The proportion of True values in the dataset

• Not enough information

Answer explanation

Think of all the False Positives

James, Amelia, and George are participating in a machine learning competition. They have to choose an algorithm for their project. Select which of the following algorithms they should consider if they want to use eager learners:

• K-nearest neighbours

• Decision trees

• Neural networks

• Linear regression

Answer explanation

Recall that K-nn is a lazy learner. At training time the algorithm simply stores the training data - i.e. no calculations/training occurs. It is not until inference time when the algorithm checks the K nearest points to the unseen datapoint in question. All calculations occur at inference time, hence being called lazy rather than eager.

Which of the following statements are True:

• Performance on the validation set can be used to see if a model is overfitting to the training data

• We cannot tell from the training performance alone if a model is overfitting or not

• Underfitting implies better generalisation to other datasets

Answer explanation

Underfitting is when the model lacks the capacity to fit the underlying pattern/trend of the data. A model that underfits a training set will perform no better on unseen data.

Scarlett is working on a machine learning project and she is worried about underfitting. Which of the following actions may cause underfitting in her model?

• Reducing the max. depth of a decision tree

• Increasing the value of K in K-nn

• Adding more layers to a neural network

• Increasing the size of the training data

• Increasing the value of K in K-means

Answer explanation

Underfitting is caused when the model lacks the capacity to fit the underlying trend/pattern of the data.

True or False:

If we use grid-search for testing different hyper-parameter values, we can use each of these results for finding the confidence interval of the model error.

• True

• False

Answer explanation

Confidence Intervals should be reported for the final model architecture. They are a prediction on how the final model will perform on unseen data. If this is calculated with a range of different models, this will clearly be an inaccurate prediction.

Which of the following algorithms will change given different random seeds:

• Neural networks

• K-nearest neighbours (K = 1, with no ties)

• Decision trees

• K-means

• Evolution Algorithms using simple tournament

Answer explanation

Think about which methods are deterministic.

Which statements below are True describing the differences between Gradient Descent, Stochastic Gradient Descent and Mini-batched Gradient Descent:

• Gradient Descent is faster to compute than Stochastic Gradient Descent

• Stochastic Gradient Descent is faster to compute than Mini-batched Gradient Descent

• There is less noise in the gradients when using Mini-batched Gradient Descent compared to Stochastic Gradient Descent

Answer explanation

Gradient descent - gradients are calculated and a step is taken based on the whole training step. This is a computationally heavy computation, as apposed to calculating the gradients and updating the parameters based on one sample. Stochastic gradient descent results in a very noising learning signal as the gradient is sensitive to the variability of each individual datapoint. The learning signal can be smoothed out by sampling groups of datapoints in mini-batch gradient descent. Here the gradient is averaged over all datapoints.

Which of the following statements about K-means are True:

• The algorithm always converges

• The algorithm always converges to a global optimum

• The algorithm doesn’t always converge

• If the algorithm does converge, it will converge to a global optimum

Answer explanation

Think about local optima. Is there a mechanism in K-means to help it get out of local minima once the algorithm has converged?

For a Gaussian Mixture Model, which of the following statements are True:

• The responsibilities r_ik for ith data point sum to 1

• The responsibilities r_ni for the ith mixture component sum to 1

• None of the above are True

Answer explanation

The responsibilities, r_nK for a particular data point, n, represents the probability of that datapoint being generated by all the K mixture components. This is a probability distribution over all mixtures and therefore has to sum to 1. A mixture, k, however, can be responsible for generating the vast majority of data points, therefore, easily having a total responsibility over all datapoints over 1.

Which of one the following is correct?

• a) sigmoid b) tanh c) ReLU

• a) tanh b) sigmoid c) Linear

• a) sigmoid b) tanh c) Linear

• a) softmax b) tanh c) ReLU

• None of the above

Answer explanation

Check definitions