house_prices.py 文件源码

def feature_agglomeration(df):
        df = df.copy()
        # Todo: find optimal number of clusters for the feature clustering
        # number_of_clusters = int(df.shape[1]/2)
        number_of_clusters = int(df.shape[1] / 1.2)
        from sklearn.cluster import FeatureAgglomeration
        agglomerated_features = FeatureAgglomeration(n_clusters=number_of_clusters)
        # mask = ~df[features].isnull()
        # mask_index = mask[mask == 1].index
        if any(tuple(df.columns == 'SalePrice')):
            # res = agglomerated_features.fit_transform(np.reshape(np.array(df[HousePrices._feature_names_num.values]
            #                                                               [mask].values),
            #                                                      df[HousePrices._feature_names_num.values][mask]
            #                                                      .shape), y=df.SalePrice.values).toarray()

            res = agglomerated_features.fit_transform(np.reshape(np.array(df.dropna().values), df.dropna()
                                                                 .shape), y=df.SalePrice.values)
        else:
            # res = agglomerated_features.fit_transform(np.reshape(np.array(df.dropna().values), df.dropna()
            #                                                      .shape))
            res = agglomerated_features.fit_transform(np.reshape(np.array(df.values), df.shape))

        # Todo: in case of adding values using df.loc[], remember mask is only possible for a single feature at a time.
        print(''.join(['labels:', str(agglomerated_features.labels_)]))
        print(''.join(['Children:', str(agglomerated_features.children_)]))
        print(''.join(['number of leaves in the hierarchical tree:', str(agglomerated_features.n_leaves_)]))
        HousePrices.dendrogram(df, number_of_clusters, agglomerated_features.labels_)
        df = pd.DataFrame(data=res)
        return df