Clustering¶
Clustering sections define how structures are grouped. The workflow task name
combines the clustering method and the selection method
(e.g. ap_center, kmeans_window).
ap¶
Affinity Propagation clustering.
[ap]
feature_name = acsf_pca
n_clusters = 100
damping = 0.5
max_iter = 200
convergence_iter = 15
max_sampling_attempts = 10
preference_range = quantile
clusters_tol = 5
max_ap_attempts = 10
max_ap_workers = 8
debug_mode = False
feature_name:str.Key in
xtal.infosaving the feature vector (e.g.acsf_pca,acsf).n_clusters:int|float.Target number of clusters. A float in (0, 1) is a fraction of the pool size. A integer is the absolute number of clusters.
damping:float| default =0.5.Damping factor for AP (between 0.5 and 1.0).
max_iter:int| default =200.Maximum AP iterations.
convergence_iter:int| default =15.Iterations with no change before declaring convergence.
max_sampling_attempts:int| default =10.Outer iterations to adjust preference and reach
n_clusters.preference_range:str|list[float]| default =quantile.Initial preference range. available options are
quantileandmean-median.clusters_tol:int|float| default =0.5.Tolerance on the final cluster count. An integer is absolute; a float is relative to
n_clusters.max_ap_attempts:int| default =10.Maximum retries per AP worker when clustering fails to converge.
max_ap_workers:int| default =8.Maximum MPI ranks used as AP workers. Additional ranks wait.
debug_mode:bool| default =False.Print per-iteration AP diagnostics.
kmeans¶
MiniBatchKMeans clustering.
[kmeans]
feature_name = acsf_pca
n_clusters = 100
save_info = False
feature_name:str.Key in
xtal.infosaving the feature vector (e.g.acsf_pca,acsf).n_clusters:int|float.Target number of clusters. A float in (0, 1) is a fraction of the pool size. A integer is the absolute number of clusters.
save_info:bool| default =False.Store per-structure distance to cluster center in
xtal.info.