python - Feature selection using tensorflow -
i using python 3.5 tensorflow 0.11.
i have dataset large number of features (>5000) , relatively small number of samples(<200). using wrapper skflow function dnnclassifier deep learning.
it seems work work classification task, want find important features large number of features.
internally, dnnclassifier seems perform feature selection(or feature extraction). there way perform feature selection tensorflow?
or, there function extract weights of features? (there function dnnclassifier.weights_, seems deprecated)
if tensorflow not support feature selection or weight information, reasonable conduct feature selection using other method(such univariate feature selection) , try deep learning?
thank help.
you can eval weights. example if variable define by
weights = tf.variable(np.ones([100,10],dtype='float32'), name=weights)
you can value @ tensorflow session
value = weights.eval();
Comments
Post a Comment