CODEWITHSUNDEEP
python-2.7pattern-matchingneural-networkfeature-detectionfeature-extraction
xor
xor

Reputation: 2698

Feature Extraction for training

I am having trouble in using feature extraction for object classification. I need to classify different types of cars but I do not know which feature extraction technique to use for training purpose.

In short How can I conclude that (Lets say) Algorithm A is good for my purpose and what features should I look for?`

Also tell me if there is any module to implement feature extraction in python.

Upvotes: 0

Views: 1013

Answers (1)

David
David

Reputation: 1136

The techniques you will use in machine learning will vary depending on your descriptors. Are they continuous or discrete ? Does your oracle estimate a continuous value, or find a discrete one ?

  • For example, if your output is continuous, you can use linear regression, SVM...
  • For discrete output, you can rely on decision trees, k-means or any clustering technique, multiple discriminant analysis.
  • Neural networks can be used for both.

Dealing with discrete descriptors can be a little trickier, you can either :

  • transform them in continuous variables
  • use regression trees, clustering trees, etc

If you have a lot of descriptors, you might need to select your descriptors during a pre processing step (feature selection) ; you can do it with forward selection for example.

You will really find a lot of stuff in R, so I would suggest you to call R from your Python code. More in this link (http://scienceoss.com/rpy-statistics-in-r-from-python/).

Cheers

Upvotes: 1

Related Questions