Sei sulla pagina 1di 8

Machine Learning Week 6 Quiz 1 (Advice

for Applying Machine Learning) Stanford


Coursera
Question 1

Answer Explanation

High This learning curve shows high error on both the training and test sets, so the algorithm
Answer Explanation

Bias bias.

Question 2
Useful Info: Since the hypothesis performs well (has low error) on the training set, it is
suffering from high variance (overfitting)
True/Fals
Answer Explanation
e

Try evaluating the hypothesis A cross validation set is useful for choosing the opt
False on a cross validation set rather parameters like the regularization parameter λ, but
than the test set. sufficient for debugging problems with the algorith

The gap in errors between training and test sugges


Try decreasing the
False problem in which the algorithm has overfit the train
regularization parameter λ.
regularization parameter will increase the overfittin

The gap in errors between training and test sugges


Try using a smaller set of problem in which the algorithm has overfit the train
True
features. feature set will ameliorate the overfitting and help
problem

The gap in errors between training and test sugges


Try increasing the problem in which the algorithm has overfit the train
True
regularization parameter λ. regularization parameter will reduce overfitting and
problem.

The gap in errors between training and test sugges


problem in which the algorithm has overfit the train
True Get more training examples
training data will increase the complexity of the tra
the variance problem.

Question 3
Useful Info: Since the hypothesis performs poorly on the training set, it is suffering from
high bias (underfitting)

True/Fals
Answer Explanation
e

The poor performance on both the training and test sets s


Try adding polynomial
True problem. Adding more complex features will increase the c
features.
hypothesis, thereby improving the fit to both the train and

False Try increasing the The poor performance on both the training and test sets s
regularization problem. Increasing the regularization parameter will allow
True/Fals
Answer Explanation
e

parameter λ. the data worse, decreasing both training and test set perfo

The poor performance on both the training and test sets s


Try using a smaller set
False problem. Using fewer features will decrease the complexity
of features.
will make the bias problem worse

The poor performance on both the training and test sets s


Try to obtain and use
True problem. Using additional features will increase the compl
additional features.
thereby improving the fit to both the train and test data.

Question 4
True/Fals
Answer Explana
e

Suppose you are training a regularized linear You should not use training
regression model.The recommended way to choose regularization parameter, as
False what value of regularization parameter λ to use is to training error by using less re
choose the value of λ which gives the lowest training value of ). But too small of a
set error. well onthe test set.

The performance of a learning algorithm on the The learning algorithm finds


True training set will typically be better than its training set error, so the perf
performance on the test set. better on the training set tha

Suppose you are training a regularized linear


The cross validation lets us fi
regression model. The recommended way to choose
setting of the regularization
True what value of regularization parameter λ to use is to
fixed model parameters learn
choose the value of λ which gives the lowest cross
set.
validation error

Suppose you are training a regularized linear


You should not use the test s
regression model. The recommended way to choose
regularization parameter, as
False what value of regularization parameter λ to use is to
artificially low value for test e
choose the value of λ which gives the lowest test
a good estimate of generaliz
set error.

This is a good split of the da


A typical split of a dataset into training, validation
bulk of the data to finding m
True and test sets might be 60% training set, 20%
training while leaving enoug
validation set, and 20% test set.
validation and estimating ge

Question 5

True/Fals
Answer Explan
e

True If a learning algorithm is suffering from high bias, only If a learning algorithm is s
True/Fals
Answer Explan
e

adding more training examples may not improve the only adding more training
test error significantly. improve the test error sign

More model parameters in


A model with more parameters is more prone to
True complexity, so it can more
overfitting and typically has higher variance.
training, increasing the ch

When debugging learning algorithms, it is useful to plot The shape of a learning cu


True a learning curve to understand if there is a high bias or of bias or variance problem
high variance problem. algorithm.

With lower training than t


If a neural network has much lower training error than
high variance. Adding mo
False test error, then adding more layers will help bring the
model complexity, making
test error down because we can fit the test set better.
worse.

Potrebbero piacerti anche