brails.imputers.knn_imputer.knn_imputer module

class brails.imputers.knn_imputer.knn_imputer.KnnImputer

Bases: Imputation

Imputes dataset based on k-nearest neighbors in the feature-agmented space. Sequentially generate inventory

Attributes:

n_pw (int):

The number of possible worlds (i.e. samples or realizations) batch_size (int):

The number of batches for sequential generation. If non-sequential, this variable is not used

gen_method (str):

Select “sequential” or “non-sequential” (one-shot). The latter is faster but does not generate the spatial correlation

seed (int):

For reproducibility

Methods:

category_in_df_to_indices(bldg_properties_df, mask)

clustering(bldg_properties_encoded, bldg_geometries_df, nbldg_per_cluster=500, seed=0)

geospatial_knn(bldg_properties_encoded, mask, bldg_geometries_df)

impute(input_inventory: AssetInventory, n_possible_worlds=1, create_correlation=True, exclude_features=[], seed=1, batch_size=50, k_nn=5) → AssetInventory

Imputate an Asset Inventory

Args:

input_inventory (AssetInventory): the inventory

Returns:

AssetInventory: a new asset inventory with missing data imputed

invetory_to_df(inventory)

Convert inventory class to df

Args:

inventory (AssetInventory):

the id of the asset

sequential_imputer(sample_dic, mp_dic, bldg_impu_subset, mask_impu_subset, bldg_inde_subset, column_names, corrMat, bldg_prel_subset, bldg_geom_subset, nbrs_G, trainY_G_list, is_category, npp, gen_method='sequential')

update_inventory(inventory, sample_dic, label_encoders, is_category)