search

akutkin / frb

0 stars 0 forks source link

More features to supply to classifiers #18

Open ipashchenko opened 8 years ago

ipashchenko commented 8 years ago

Currently only 9 and only 2-3 of them are informative (for GBC). Should i use feature selection or let the algo decide what are the most informative ones (like GBC)?

ipashchenko commented 8 years ago

Commit bb869a3d7421936c0a44a9a611067da627ede49a uses this features: prop.area, gg.amplitude.value, abs(gg.x_stddev.value), abs(gg.y_stddev.value), abs(gg.theta.value), abs(gg.x_stddev.value/gg.y_stddev.value), prop.extent, abs(gg.amplitude/prop.mean_intensity), prop.solidity, prop.major_axis_length, prop.minor_axis_length, prop.perimeter, prop.max_intensity, prop.mean_intensity, prop.weighted_moments_hu[0], prop.weighted_moments_hu[1], prop.weighted_moments_hu[2], prop.orientation, prop.inertia_tensor_eigvals[0], prop.inertia_tensor_eigvals[1], prop.filled_area, prop.euler_number, prop.eccentricity, prop.convex_area Only 3-4 of them are informative as GBC shows. EDIT:

Feature importance : [ 0.          0.18400372  0.          0.01957175  0.          0.          0.
  0.11580235  0.          0.          0.          0.          0.21075891
  0.23584882  0.06206432  0.04069084  0.          0.          0.          0.
  0.          0.          0.13125928  0.        ]