PCA method for Feature Selection - How do I resolve the raise Exception("Data must be 1-dimensional" error)?

Question

I am trying to implement the PCA method for Feature Selection from the following functions:

#Função que permitirá rankear as features mais importantes em um barhplot
def ranks_PCA (x_train, y_train, features_train, RESULT_PATH='Results'):
    print("\nMétodo PCA")

    pca = PCA(n_components=58)
    pca.fit_transform(x_train)

    imp_array = np.array(pca.components_)
    imp_order = imp_array.argsort()
    ranks = imp_order.argsort()

    # Plot PCA
    imp = pd.Series(pca.components_, index=x_train.columns)
    imp = imp.sort_values()

    imp.plot(kind="barh")
    plt.xlabel("Importance")
    plt.ylabel("Features")
    plt.title("Feature importance using PCA")
    # plt.show()
    plt.savefig(RESULT_PATH + '/ranks_DT.png', bbox_inches='tight')

    return ranks

---

#Função para predição das features dos dados de teste
def predict_PCA(x_test_sel, k_vetor, y_train):
    model = decomposition.PCA()
    model.fit(k_vetor, y_train)
    y_predict = model.predict(x_test_sel)
    return(y_predict)

---

#Função que calcula o ranking dos dados de treinamento
ranks4 = frk.ranks_PCA(x_train, y_train, features_train, RESULT_PATH)

I have doubts whether this implementation is correct to get more important Features. When trying to run this code, I get the following error:

Traceback (Most recent call last): File "feat_test.py", line 235, in 'Results/Pdbbind2018_f58_delta_pkd') File "feat_test.py", line 78, in run_experiment ranks4 = frk.ranks_PCA(x_train, y_train, features_train, RESULT_PATH) File "C: Users Patricia Desktop VT-58 - Copy Feature-importance feature_rank_ ensemble Scripts feature_ranks.py", line 121, in ranks_PCA imp = pd.Series(pca.components_, index=x_train.Columns) File "C: Users Patricia Desktop VT-58 - Copy Feature-importance feature_rank_ ensemble env lib site-Packages pandas core series.py", line 305, in init data = sanitize_array(data, index, dtype, copy, raise_cast_failure=True) File "C: Users Patricia Desktop VT-58 - Copy Feature-importance feature_rank_ ensemble env lib site-Packages pandas core Construction.py", line 482, in saniti ze_array raise Exception("Data must be 1-dimensional")

Could someone help me?

Answer 1

The attribute Components_ of sklearn.decomposition.PCA returns an array with Shape equal to (n_components, n_features). In your case (58, x.Shape[0]).

You are using this array to try to create an object of the type pandas. Series, accepting only 1 dimension arrays.

The mistake is in:

imp = pd.Series(pca.components_, index=x_train.columns)

If you want to plot only one component, you must choose it or change the number of PCA components to 1.