selector-model-devi.yml

# This is the configuration file for select stage with model deviation based selector.


workflow:
  select:
    model_devi:
        # The structure
        # whose max_devi_f is less than f_trust_lo will be considered as good.
        # whose max_devi_f is larger than f_trust_hi will be considered as poor.
        # whose max_devi_f is between f_trust_lo and f_trust_hi will be considered as decent.
        # Only the decent structures will be used selected as new training set.
        f_trust_lo: 0.4
        f_trust_hi: 0.6

        # Optional, the quantile of model_devi score to select the structure
        # for next round of exploration
        new_explore_system_q: 0.25

        # Optional, limit max decent per trajectory
        # Default: -1 (no limit)
        max_decent_per_traj: 1

        # Optional, number of workers for parallel computing, set it to 1 for debug.
        # Default: 4
        workers: 1

        # Optional, filter structures by condition defined by lambda string
        # The only input argument of lambda function is `ase.Atoms`
        # The following local variable can be used in the expression:
        # ssw_energy_quantile, ssw_energy_max, ssw_energy_min
        # e.g: "lambda x: x.info['ssw_energy'] < ssw_energy_quantile(0.25)"
        screening_fn:

        # Optional, select most dissimilar structures by clustering.
        # If it is None, this stage will be skipped.
        #
        # The process of this stage is as follow:
        #
        # 1. Get global descriptors of structures, use methods like SOAP, ACSF, etc.
        # 2. Reduce the dimension of global descriptors, use methods like PCA, t-SNE, etc.
        # 3. Cluster the reduced global descriptors, use methods like DBSCAN, etc.
        # 4. Select the most dissimilar structures from each cluster.
        #
        # You can set it to {}  if you want to use default options.
        # or you can specify the params directly as the follow
        asap_options:

          # Optional, limit the number of structures selected from each cluster.
          # Default: 1
          # Set it to 0 if you want keep all structures.
          limit_per_cluster: 1

          # Optional, whether to sort the structures by ssw_energy.
          # Default: false
          sort_by_ssw_energy: true

          # Optional, specify the method to get global descriptors.
          # Only one of soap, acsf and cm need to be set.
          # Default: soap
          # Remember to remove the unused ones for the configuration file.
          descriptor:
            # Optional, use SOAP descriptor.
            soap:
              # Optional, the preset configuration provided by asaplib,
              # includes: smart, minimal, longrangeb
              # You can skip the other params if preset is not None
              preset:

              # Use can also specify the params directly.
              # The following params will take effect only when preset is None.
              # those params following the convention of dscribe
              # https://singroup.github.io/dscribe/latest/tutorials/descriptors/soap.html
              r_cut: 3.5
              n_max: 6
              l_max: 6
              sigma: 0.5

              crossover: false
              rbf: gto

              reducer_type: average
              element_wise: false
              zeta: 1

            # Optional, use ACSF descriptor.
            acsf:
              # Optional, the preset configuration provided by asaplib,
              # includes: smart, minimal, longrangeb
              # You can skip the other params if preset is not None

              preset:
              # those params following the convention of dscribe
              # https://singroup.github.io/dscribe/latest/tutorials/descriptors/acsf.html
              r_cut: 3.5

              reducer_type: average
              element_wise: false
              zeta: 1

            # Optional, use CM descriptor.
            # No params need to be specified, just set it to {} if you want to use it.
            cm: {}

          # Optional, specify the method to reduce the dimension of global descriptors.
          # Only one of pca, tsne, umap and autoencoder need to be set.
          # Default: pca
          # The params are used by the asaplib
          # Remember to remove the unused ones for the configuration file.
          dim_reducer:
            # Optional, use PCA and specify the params.
            pca:
              type: PCA
              parameter:
                n_components: 3
                scalecenter: true

          # Optional, specify the method to cluster the reduced global descriptors.
          # Only one of dbscan and laio_db need to be set.
          # Default: dbscan
          # Remember to remove the unused ones for the configuration file.
          cluster:

            # Optional, use DBSCAN and specify the params.
            # Set it to {} if you want to use default params.
            # Or you can specify the params directly as the follow
            dbscan:
              eps:
              min_samples: 2

            # Optional, use LaioDB, no params need to be specified.
            laiodb: {}