Challanges in Machine Learning Flashcards
1
Q
How can the problem of irrelevant features be solved?
A
- Remove features
- Combine features
- Gather data to create new features
2
Q
How can the problem of overfitting be fixed?
A
- Gather more data
- Simplify the model
- Reduce training set noise
3
Q
What are some ways an ML model can be simplified?
A
- Using a simpler algorithm
- Reducing the number of model parameters being used
- Regularizing the model
4
Q
How can the problem of under-fitting be solved?
A
- Use a more complex model
- Feature engineering
- Reduce the number of learning constraints
5
Q
How can error in the training set be rectified?
A
- Remove outliers
- Find missing features
6
Q
How is the problem of data mismatch solved?
A
Separating the mismatched data into a seperate data set.
7
Q
How can you use the train-dev set to discover overfitting?
A
If the model does well on the test set, but not the train-dev set.
8
Q
How you use the train-dev set to discover mismatched data in the training set?
A
Model does well on training and train-dev set, but not the validation set.
