Configuration templates¶
Here we present some configuration templates that we used during our experiments. Used on the .tif
sequence above, they produce good results. In order for you to see what we call “good” results, they are shown above too for each configuration template.
The initial sequence.
The configuration templates can be found in the pyamnesia/config_templates/
directory.
Clustering templates¶
Skeleton pixels clustering¶
- file:
config_clustering_pixel.yaml
- context: clustering the traces of skeleton pixels from the
.tif
sequence. - specs:
projection_substructure
is set to'skeleton'
andprojection_method
to'pixel'
. t-SNE perplexity is relatively high, because we expect to find mesoscopic clusters composed of dozens of pixels. - results:
Clustering plot, cluster overlay and cluster traces.
Skeleton branches clustering¶
- file:
config_clustering_branch.yaml
- context: clustering the traces of skeleton branches (or mesoscopic clusters from the pixel clustering) from the
.tif
sequence. - specs:
projection_substructure
is set to'skeleton'
andprojection_method
to'branch'
. t-SNE perplexity is lower, because we expect to find macroscopic clusters composed of a few branches (or mesoscopic clusters). - results:
Clustering plot, cluster overlay and cluster traces.
Active pixels clustering¶
- file:
config_clustering_active.yaml
- context: clustering the traces of active pixels from the
.tif
sequence. - specs:
projection_substructure
is set to'active'
andprojection_method
to'pixel'
. t-SNE perplexity is lower, because we expect to find macroscopic clusters composed of a few branches (or mesoscopic clusters). - results:
Clustering plot, cluster overlay and cluster traces.
Factorization templates¶
NMF¶
- file:
config_factorization_nmf.yaml
- context: performing a NMF on the
.tif
sequence. - specs:
element_normalization_method
is set tonull
so that the input of the NMF is nonnegative. The number of components is set to50
but can be adapted. - results:
Two components and the skewness histogram (open image in new tab for a better view).
PCA and ICA¶
- file:
config_factorization_pca_ica.yaml
- context: performing a PCA and an ICA on the
.tif
sequence. - specs:
element_normalization_method
is set to'z-score'
. The number of components is set to50
but can be adapted. - results:
Two components and the skewness histogram (open image in new tab for a better view).