jni / ray

Automatic segmentation of electron microscopy volumes
Other
31 stars 10 forks source link

Run unique() on training data after compiling #23

Closed jni closed 13 years ago

jni commented 13 years ago

Training proceeds by multiple iterations of random merging in a volume. During training some node pairs may be merged that were encountered in previous iterations, resulting in duplicate entries in the training data, which could adversely affect performance.

We should try ensuring that each data entry is unique and see how classifier performance is affected.

jni commented 13 years ago

Ryan ran this and saw a 16% reduction in training set size, but no effect on classifier performance.