Name

dnn_fe - Extract features of DNN activation.

The dnn_fe using methods to reduct feature dimension, clustering several main components of the DNN high dimensional activation as required

Synopsis

dnn_fe -act Activation [-layer Layer [Layer …]] [-chn Channel [Channel …]] [-dmask DnnMask] -meth Method [-axis Axis] -out Output

Arguments

Required Arguments

Argument	Discription
act	Path of a .act.h5 file which contains activation information to extract features.
meth	Method of feature extraction.Enter two parameters in order (‘Method’, ‘N_feature’).The first means the method choosing from (‘hist’, ‘psd’, ‘pca’).The second is used to specify the number of features we will use.
out	Output path with a suffix as .act.h5.

Optional Arguments

Argument	Discription
layer	Name of the target layer(s).Default is all.E.g., ‘conv1’ represents the first convolution layer, and ‘fc1’ represents the first full connection layer.
chn	Index of target channel(s).Default is all.Channel index starts from 1.
dmask	Path of a .dmask.csv file in which detailed information of neuron(s) of interest in DNN is specified.Argument layer/chn and dmask are mutually exclusive. Provide only one of them if needed.
axis	Axis for feature extraction.Default is extracting features from the whole layer.chn: axis in chn; row_col: axis in specific row_col unit in a feature map;

Outputs

An .act.h5 file containing the feature extracted activation that can be read and saved with the module dnnbrain.io.fileio.ActivationFile.

Examples

These examples demonstrate the activation feature extraction function. Activation provided by test.act.h5 was extracted feature and finally saved in the dnn_fe.act.h5 file.

# Asserting target layers using the -layer argument
dnn_fe -act ./test.act.h5 -layer conv1 conv5_relu fc2_relu -stim ./test.stim.csv -meth pca 3 -axis chn -out ./dnn_fe1.act.h5

# Asserting target layers using the -dmask argument
dnn_fe -act ./test.act.h5 -dmask ./test.dmask.csv -stim ./test.stim.csv -meth pca 3 -axis chn -out ./dnn_fe2.act.h5

# Not asserting layers which means target layers and chns are all
dnn_fe -act ./test.act.h5 -stim ./test.stim.csv -meth pca 3 -axis chn -out ./dnn_fe3.act.h5