# The SVM baseline for SEMAINE import shared_defs_SEMAINE import data_preparation import numpy import mlp (all_aus, train_recs, devel_recs, SEMAINE_dir, hog_data_dir) = shared_defs_SEMAINE.shared_defs() pca_loc = "../pca_generation/generic_face_rigid" # load the training and testing data for the current fold [train_samples, train_labels, valid_samples, valid_labels, raw_valid, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_SEMAINE_dynamic(train_recs, devel_recs, all_aus, SEMAINE_dir, hog_data_dir, pca_loc, scale=False) import validation_helpers train_fn = mlp.train_mlp_probe test_fn = mlp.test_mlp_class hyperparams = { 'batch_size': [100], 'learning_rate': [0.2], 'lambda_reg': [0.0001], 'num_hidden': [100, 250, 400], 'n_epochs': 1000, 'error_func': 'cross_ent', 'final_layer': ['tanh', 'sigmoid'], 'validate_params': ["batch_size", "learning_rate", "lambda_reg", 'num_hidden']}
# The SVM baseline for BO4D import shared_defs_SEMAINE import data_preparation (all_aus, train_recs, devel_recs, SEMAINE_dir, hog_data_dir) = shared_defs_SEMAINE.shared_defs() pca_loc = "../pca_generation/generic_face_rigid" f = open("./trained/SEMAINE_train_dynamic_lin_svm_geometry.txt", 'w') for au in all_aus: hyperparams = {"C": [0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000], "validate_params": ["C"]} # load the training and testing data for the current fold [train_samples, train_labels, valid_samples, valid_labels, raw_valid, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_SEMAINE_dynamic(train_recs, devel_recs, [au], SEMAINE_dir, hog_data_dir, pca_loc, geometry=True) import linear_SVM import validation_helpers train_fn = linear_SVM.train_SVM test_fn = linear_SVM.test_SVM # Cross-validate here best_params, all_params = validation_helpers.validate_grid_search(train_fn, test_fn, False, train_samples, train_labels, valid_samples, valid_labels, hyperparams) model = train_fn(train_labels, train_samples, best_params)
# The SVM baseline for SEMAINE import shared_defs_SEMAINE import data_preparation import logistic_regression (all_aus, train_recs, devel_recs, SEMAINE_dir, hog_data_dir) = shared_defs_SEMAINE.shared_defs() pca_loc = "../pca_generation/generic_face_rigid" f = open("./trained/SEMAINE_train_static_log_reg.txt", 'w') for au in all_aus: # load the training and testing data for the current fold [train_samples, train_labels, valid_samples, valid_labels, raw_valid, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_SEMAINE(train_recs, devel_recs, [au], SEMAINE_dir, hog_data_dir, pca_loc) import validation_helpers train_fn = logistic_regression.train_log_reg test_fn = logistic_regression.test_log_reg hyperparams = { 'batch_size': [100], 'learning_rate': [0.05, 0.1, 0.2], 'lambda_reg': [0, 0.05, 0.1, 0.5], 'n_epochs': 100, 'validate_params': ["batch_size", "learning_rate", "lambda_reg"] }
import shared_defs_DISFA import data_preparation import numpy import mlp pca_loc = "../pca_generation/generic_face_rigid" (all_aus_bp4d, train_recs, devel_recs, BP4D_dir, hog_data_dir) = shared_defs_BP4D.shared_defs() # load the training and testing data for the current fold [train_samples_bp4d, train_labels_bp4d, valid_samples_bp4d, valid_labels_bp4d, _, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_BP4D(train_recs, devel_recs, all_aus_bp4d, BP4D_dir, hog_data_dir, pca_loc, geometry=True) (all_aus_semaine, train_recs, devel_recs, semaine_dir, hog_data_dir) = shared_defs_SEMAINE.shared_defs() # load the training and testing data for the current fold [train_samples_semaine, train_labels_semaine, valid_samples_semaine, valid_labels_semaine, _, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_SEMAINE(train_recs, devel_recs, all_aus_semaine, semaine_dir, hog_data_dir, pca_loc, geometry=True) (all_aus_disfa, train_recs, disfa_dir, hog_data_dir) = shared_defs_DISFA.shared_defs() devel_recs = train_recs[0:1] [train_samples_disfa, train_labels_disfa, _, _, _, PC, means, scaling] = \ data_preparation.Prepare_HOG_AU_data_generic_DISFA(train_recs, devel_recs, all_aus_disfa, disfa_dir, hog_data_dir, pca_loc, geometry=True) # Binarise disfa labels train_labels_disfa[train_labels_disfa > 1] = 1