Conversation
avplatonov
left a comment
avplatonov
left a comment
There was a problem hiding this comment.
@ilantukh Thanks for your contribution!
Please review my comments.
| private final Preprocessor<K, V> basePreprocessor; | ||
|
|
||
| /** DCT type, default is 2 */ | ||
| private final int type; |
There was a problem hiding this comment.
Why did you use int as old-c-style-enum? I think we should replace int to enum.
|
|
||
| IgniteCache<Integer, Vector> persons = ignite.createCache(cacheConfiguration); | ||
|
|
||
| persons.put(1, new DenseVector(new Serializable[]{"Mike", 10, 20, 42})); |
There was a problem hiding this comment.
Such example looks a little weird because you apply DCT to irrelevant data (or data of unknown nature). I think in this example will be great to use vectors with time-series. Maybe it can be data from fake-sensors and we should classify who is broken between them by values from sensors. In this case, a vector will be an array of doubles with fixed size (sensor values per hour) and a label will be 0 or 1 where 0 denotes "good sensor" and 1 denotes "broken sensor".
| * @param <K> Type of a key in {@code upstream} data. | ||
| * @param <V> Type of a value in {@code upstream} data. | ||
| */ | ||
| public class DiscreteCosinePreprocessor<K, V> implements Preprocessor<K, V> { |
There was a problem hiding this comment.
Because of DCT processor should be applied to sequential data and vectors with features data could contain data of different nature I think it will be great if you provide an ability to pass filter of features to this preprocessor. In this case preprocessor will process just these filtered features and replace old values by new preprocessed values.
2 participants
No description provided.