def blogcatalog_test_scenario(deepwalk_path): y_path = '../../local_resources/blogcatalog/y.p' x_path = '../../local_resources/blogcatalog/X.p' target = utils.read_target(y_path) x, y = utils.read_data(x_path, y_path, threshold=0) names = [['deepwalk'], ['logistic']] x_deepwalk = pd.read_csv(deepwalk_path, index_col=0) # all_features = np.concatenate((x.toarray(), x_deepwalk), axis=1) X = [x_deepwalk.values, normalize(x, axis=0)] n_folds = 10 results = run_detectors.run_all_datasets(X, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/karate/deepwalk_macro_pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('../../results/karate/deepwalk_micro_pvalues' + utils.get_timestamp() + '.csv') print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/karate/deepwalk_macro' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/karate/deepwalk_micro' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def blogcatalog_deepwalk_node2vec(): paths = [ 'local_resources/blogcatalog/blogcatalog128.emd', 'local_resources/blogcatalog/blogcatalog_p025_q025_d128.emd' ] names = [['logistic_p1_q1'], ['logistic_p025_q025']] y_path = 'local_resources/blogcatalog/y.p' detectors = [classifiers_embedded_128, classifiers_embedded_128] sizes = [128, 128] X, y = read_embeddings(paths, y_path, sizes) n_folds = 5 results = run_all_datasets(X, y, names, detectors, n_folds) all_results = utils.merge_results(results) results = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = 'results/blogcatalog/macro_deepwalk_node2vec' + utils.get_timestamp( ) + '.csv' micro_path = 'results/blogcatalog/micro_deepwalk_node2vec' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def test_embeddings(): feature_path = '../local_resources/features_1in10000.tsv' rf_features = pd.read_csv(feature_path, sep='\t', index_col=0) emd = pd.read_csv('../local_resources/hyperbolic_embeddings/tf_test1.csv', header=None, index_col=0, skiprows=1, sep=" ") features, y = utils.get_classification_xy(rf_features) features = features.loc[emd.index, :] y = y.loc[emd.index].values names = np.array([['RF just emd']]) n_folds = 10 classifiers = [ RandomForestClassifier(max_depth=2, n_estimators=50, bootstrap=True, criterion='entropy', max_features=0.1, n_jobs=1) ] results = run_all_datasets([emd.values], y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = 'tf_testing_1in10000' + utils.get_timestamp() + '.csv' micro_path = 'tf_micro_1in10000' + utils.get_timestamp() + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True) assert results[0]['mean'].values > 0.6
def blogcatalog_121_scenario(embedding_path): target_path = '../../local_resources/blogcatalog_121_sample/y.p' feature_path = '../../local_resources/blogcatalog_121_sample/X.p' hyperbolic = pd.read_csv(embedding_path, index_col=0).values paths = ['../../local_resources/blogcatalog_121_sample/blogcatalog2.emd'] sizes = [128] [deepwalk], y = read_embeddings(paths, target_path, sizes) names = [['logistic'], ['deepwalk'], ['hyp embedding']] x = utils.read_pickle(feature_path) # y = utils.read_pickle(target_path) X = [x, deepwalk, hyperbolic] n_folds = 2 results = run_all_datasets(X, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = '../../results/blogcatalog_121_sample/macro' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/blogcatalog_121_sample/micro' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def karate_scenario(): deepwalk_path = 'local_resources/zachary_karate/size8_walks1_len10.emd' y_path = 'local_resources/zachary_karate/y.p' x_path = 'local_resources/zachary_karate/X.p' target = utils.read_target(y_path) x, y = utils.read_data(x_path, y_path, threshold=0) names = [['logistic'], ['deepwalk']] x_deepwalk = utils.read_embedding(deepwalk_path, target) # all_features = np.concatenate((x.toarray(), x_deepwalk), axis=1) X = [x_deepwalk, normalize(x, axis=0)] n_folds = 2 results = run_all_datasets(X, y, names, classifiers, n_folds) all_results = utils.merge_results(results) results, tests = utils.stats_test(all_results) tests[0].to_csv('results/karate/deepwalk_macro_pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('results/karate/deepwalk_micro_pvalues' + utils.get_timestamp() + '.csv') print 'macro', results[0] print 'micro', results[1] macro_path = 'results/karate/deepwalk_macro' + utils.get_timestamp( ) + '.csv' micro_path = 'results/karate/deepwalk_micro' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def tf_train100000_emd_scenario(): scaler = StandardScaler() feature_path = '../../local_resources/features_train100000.tsv' # feature_path = '../../local_resources/features_train100000.tsv' rf_features = pd.read_csv(feature_path, sep='\t', index_col=0) del rf_features.index.name emd = pd.read_csv('../../local_results/tf_train_100000.emd', header=None, index_col=0, skiprows=1, sep=" ") # emd = pd.read_csv('../../local_results/tf_train_100000.emd', header=None, index_col=0, skiprows=1, sep=" ") features, y = utils.get_classification_xy(feature_path, emd) all_feat = features.join(emd) X1 = features.values.astype(np.float) X1 = scaler.fit_transform(X1) X2 = all_feat.values.astype(np.float) X2 = scaler.fit_transform(X2) names = np.array([['L2 without emd', 'L1 without emd', 'RF without emd'], ['L2 with emd', 'L1 with emd', 'RF with emd'], ['L2 just emd', 'L1 just emd', 'RF just emd']]) n_folds = 10 results = run_all_datasets([X1, X2, emd.values], y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = '../../results/neural/tf_macro_train100000' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/neural/tf_micro_train100000' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def batch_size_scenario(): """ Generate embeddings using different batch sizes for the ~1000 vertex polblogs network :return: """ import visualisation s = datetime.datetime.now() y_path = '../../local_resources/political_blogs/y.p' x_path = '../../local_resources/political_blogs/X.p' y = utils.read_pickle(y_path) log_path = '../../local_resources/tf_logs/polblogs/' walk_path = '../../local_resources/political_blogs/walks_n1_l10.csv' size = 2 # dimensionality of the embedding batch_sizes = [1, 2, 4, 8, 16, 32, 64, 128] embeddings = [] for batch_size in batch_sizes: params = Params(walk_path, batch_size=batch_size, embedding_size=size, neg_samples=5, skip_window=5, num_pairs=1500, statistics_interval=10.0, initial_learning_rate=0.1, save_path=log_path, epochs=5, concurrent_steps=4) path = '../../local_resources/political_blogs/embeddings/Win_batch_{}_{}.csv'.format( batch_size, utils.get_timestamp()) embedding_in, embedding_out = HCE.main(params) visualisation.plot_poincare_embedding(embedding_in, y, '../../results/political_blogs/figs/poincare_polar_Win_batch_{}_{}.pdf'.format( batch_size, utils.get_timestamp())) visualisation.plot_poincare_embedding(embedding_out, y, '../../results/political_blogs/figs/poincare_polar_Wout_batch_{}_{}.pdf'.format( batch_size, utils.get_timestamp())) df_in = pd.DataFrame(data=embedding_in, index=np.arange(embedding_in.shape[0])) df_in.to_csv(path, sep=',') df_out = pd.DataFrame(data=embedding_out, index=np.arange(embedding_out.shape[0])) df_out.to_csv( '../../local_resources/political_blogs/embeddings/Wout_batch_{}_{}.csv'.format( batch_size, utils.get_timestamp()), sep=',') print('political blogs embedding generated in: ', datetime.datetime.now() - s) embeddings.append(embedding_in) x, y = utils.read_data(x_path, y_path, threshold=0) names = [[str(batch_size)] for batch_size in batch_sizes] n_folds = 10 results = run_detectors.run_all_datasets(embeddings, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/political_blogs/batch_size_pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('../../results/political_blogs/batch_size_pvalues' + utils.get_timestamp() + '.csv') print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/political_blogs/batch_size_macro' + utils.get_timestamp() + '.csv' micro_path = '../../results/political_blogs/batch_size_micro' + utils.get_timestamp() + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True) return path
def gensim_1in10000_emd_scenario(): scaler = StandardScaler() feature_path = '../../local_resources/features_1in10000.tsv' rf_features = pd.read_csv(feature_path, sep='\t', index_col=0) emd = pd.read_csv('../../local_results/customer.emd', header=None, index_col=0, skiprows=1, sep=" ") features, y = utils.get_classification_xy(rf_features) # select only the data points that we have embeddings for features = features.loc[emd.index, :] y = y.loc[emd.index].values all_feat = features.join(emd, how='inner') print 'input features shape', all_feat.shape X1 = features.values.astype(np.float) X1 = scaler.fit_transform(X1) X2 = all_feat.values.astype(np.float) X2 = scaler.fit_transform(X2) # names = np.array( # [['L2 without emd'], ['L2 with emd']]) names = np.array([['L2 without emd'], ['L2 with emd'], ['L2 just emd']]) # names = np.array( # [['L2 without emd', 'L1 without emd', 'RF without emd'], ['L2 with emd', 'L1 with emd', 'RF with emd'], # ['L2 just emd', 'L1 just emd', 'RF just emd']]) # names = np.array([['without MF'], ['with MF']]) n_folds = 5 # np.random.seed(42) clf = LogisticRegression(multi_class='ovr', penalty='l2', solver='liblinear', n_jobs=1, max_iter=1000, C=0.005) df = run_repetitions([X1, X2, emd.values], y, clf, names, reps=10) print df # results = run_all_datasets([X1, X2], y, names, [clf], n_folds) results = run_all_datasets([X1, X2, emd.values], y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = '../../results/neural/gensim_1in10000' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/neural/gensim_1in10000' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def karate_deepwalk_grid_scenario(): """ evaluates a grid of embeddings at different sizes, walk lengths and walks per vertex for the karate network. Trying to understand why the DeepWalk performance was so poor. :return: """ import os y_path = '../../local_resources/karate/y.p' x_path = '../../local_resources/karate/X.p' target = utils.read_target(y_path) x, y = utils.read_data(x_path, y_path, threshold=0) folder = '../../local_resources/karate/gridsearch/' names = [[elem] for elem in os.listdir(folder)] embeddings = [] for name in names: emb = pd.read_csv(folder + name[0], header=None, index_col=0, skiprows=1, sep=" ") emb.sort_index(inplace=True) embeddings.append(emb.values) names.append(['hyperbolic']) hyp_path = '../../local_resources/karate/embeddings/Win_20170808-185202.csv' hyp_emb = pd.read_csv(hyp_path, index_col=0) embeddings.append(hyp_emb.values) n_folds = 10 results = run_detectors.run_all_datasets(embeddings, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/karate/pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('../../results/karate/pvalues' + utils.get_timestamp() + '.csv') print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/karate/macro' + utils.get_timestamp() + '.csv' micro_path = '../../results/karate/micro' + utils.get_timestamp() + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def MF_scenario(): scaler = StandardScaler() feature_path = '../../local_resources/features_1in10000.tsv' rf_features = pd.read_csv(feature_path, sep='\t', index_col=0) del rf_features.index.name emd = pd.read_csv('../../local_resources/roberto_emd.csv', header=None, index_col=0) del emd.index.name # emd = reduce_embedding(emd) # filter the features by customer ID temp = rf_features.join(emd[1], how='inner') features = temp.drop(1, axis=1) # extract the churn target labels print 'class distribution', features['target_churned'].value_counts() y = features['target_churned'].values.astype(int) # remove the labels features = features.ix[:, :-4] # encode the categoricals features['shippingCountry'] = utils.convert_to_other( features['shippingCountry'], pct=0.05, label='Other') features = pd.get_dummies(features, columns=['shippingCountry', 'gender']) all_feat = features.join(emd) X1 = features.values.astype(np.float) X1 = scaler.fit_transform(X1) X2 = all_feat.values.astype(np.float) X2 = scaler.fit_transform(X2) names = np.array([['L2 without MF', 'L1 without MF', 'RF without MF'], ['L2 with MF', 'L1 with MF', 'RF with MF'], ['L2 just MF', 'L1 just MF', 'RF just MF']]) # names = np.array([['without MF'], ['with MF']]) n_folds = 20 results = run_all_datasets([X1, X2, emd.values], y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = '../../results/MF/macro_1of100000no_cat' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/MF/micro_1of100000no_cat' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def karate_results(embeddings, names, n_reps, train_size): deepwalk_path = '../../local_resources/zachary_karate/size8_walks1_len10.emd' y_path = '../../local_resources/zachary_karate/y.p' x_path = '../../local_resources/zachary_karate/X.p' target = utils.read_target(y_path) x, y = utils.read_data(x_path, y_path, threshold=0) # names = [['embedding'], ['logistic']] names.append(['logistics']) # x_deepwalk = utils.read_embedding(deepwalk_path, target) # all_features = np.concatenate((x.toarray(), x_deepwalk), axis=1) # X = [normalize(embedding, axis=0), normalize(x, axis=0)] X = embeddings + [normalize(x, axis=0)] # names = ['embedding'] # X = embedding results = [] for exp in zip(X, names): tmp = run_detectors.run_experiments(exp[0], y, exp[1], classifiers, n_reps, train_size) results.append(tmp) all_results = utils.merge_results(results, n_reps) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/karate/tf_macro_pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('../../results/karate/tf_micro_pvalues' + utils.get_timestamp() + '.csv') print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/karate/tf_macro' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/karate/tf_micro' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True) return results
def political_blogs_scenario(embedding_path): # deepwalk_path = '../../local_resources/hyperbolic_embeddings/tf_test1.csv' y_path = '../../local_resources/political_blogs/y.p' x_path = '../../local_resources/political_blogs/X.p' sizes = [2, 4, 8, 16, 32, 64, 128] deepwalk_embeddings = [] deepwalk_names = [] dwpath = '../../local_resources/political_blogs/political_blogs' for size in sizes: path = dwpath + str(size) + '.emd' de = pd.read_csv(path, header=None, index_col=0, skiprows=1, sep=" ") de.sort_index(inplace=True) deepwalk_embeddings.append(de.values) deepwalk_names.append(['deepwalk' + str(size)]) x, y = utils.read_data(x_path, y_path, threshold=0) names = [['hyperbolic'], ['logistic']] names = deepwalk_names + names embedding = pd.read_csv(embedding_path, index_col=0) # all_features = np.concatenate((x.toarray(), x_deepwalk), axis=1) X = deepwalk_embeddings + [embedding.values, normalize(x, axis=0)] n_folds = 10 results = run_detectors.run_all_datasets(X, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/political_blogs/pvalues' + utils.get_timestamp() + '.csv') tests[1].to_csv('../../results/political_blogs/pvalues' + utils.get_timestamp() + '.csv') print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/political_blogs/macro' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/political_blogs/micro' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def gensim_1in10000_debug_scenario(): scaler = StandardScaler() feature_path = '../../local_resources/features_1in10000.tsv' rf_features = pd.read_csv(feature_path, sep='\t', index_col=0) del rf_features.index.name print 'input features shape', rf_features.shape emd = pd.read_csv('../../local_results/customer.emd', header=None, index_col=0, skiprows=1, sep=" ") print 'input embedding shape', rf_features.shape # emd = pd.read_csv('../../local_results/customer.emd', header=None, index_col=0, skiprows=1, sep=" ") features, y = utils.get_classification_xy(feature_path, emd) assert len(features) == features.index.values.unique() # all_feat = features.join(emd, how='inner') all_feat = features.join(emd) X1 = features.values.astype(np.float) X1 = scaler.fit_transform(X1) X2 = all_feat.values.astype(np.float) X2 = scaler.fit_transform(X2) names = np.array([['L2 without emd', 'L1 without emd', 'RF without emd'], ['L2 with emd', 'L1 with emd', 'RF with emd'], ['L2 just emd', 'L1 just emd', 'RF just emd']]) # names = np.array([['without MF'], ['with MF']]) n_folds = 10 results = run_all_datasets([X1, X2, emd.values], y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) print 'macro', results[0] print 'micro', results[1] macro_path = '../../results/neural/gensim_1in10000' + utils.get_timestamp( ) + '.csv' micro_path = '../../results/neural/gensim_1in10000' + utils.get_timestamp( ) + '.csv' results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)
def run_scenario(folder, embedding_path): y_path = '../../local_resources/{}/y.p'.format(folder) x_path = '../../local_resources/{}/X.p'.format(folder) sizes = [2, 4, 8, 16, 32, 64, 128] deepwalk_embeddings = [] deepwalk_names = [] dwpath = '../../local_resources/{0}/{1}'.format(folder, folder) for size in sizes: path = dwpath + str(size) + '.emd' de = pd.read_csv(path, header=None, index_col=0, skiprows=1, sep=" ") de.sort_index(inplace=True) deepwalk_embeddings.append(de.values) deepwalk_names.append(['deepwalk' + str(size)]) x, y = utils.read_data(x_path, y_path, threshold=0) names = [['hyperbolic'], ['logistic']] names = deepwalk_names + names embedding = pd.read_csv(embedding_path, index_col=0) X = deepwalk_embeddings + [embedding.values, normalize(x, axis=0)] n_folds = 10 results = run_detectors.run_all_datasets(X, y, names, classifiers, n_folds) all_results = utils.merge_results(results, n_folds) results, tests = utils.stats_test(all_results) tests[0].to_csv('../../results/{0}/pvalues{1}.csv'.format( folder, utils.get_timestamp())) tests[1].to_csv('../../results/{0}/pvalues{1}.csv'.format( folder, utils.get_timestamp())) print('macro', results[0]) print('micro', results[1]) macro_path = '../../results/{0}/macro{1}.csv'.format( folder, utils.get_timestamp()) micro_path = '../../results/{0}/micro{1}.csv'.format( folder, utils.get_timestamp()) results[0].to_csv(macro_path, index=True) results[1].to_csv(micro_path, index=True)