all_test_x, pd.read_csv('../predict_location/pl_L1_ET_vA1_submission.csv').values[:, 3:]), axis=1) dataset = { 'train_x': all_train_x, 'train_y': all_train_y, 'train_seq': train_seq, 'test_x': all_test_x } # Add past data dataset = { 'train_x': np.c_[all_train_x, fd.get_past_data(all_train_x, train_seq, 1, 0), fd.get_past_data(all_train_x, train_seq, 2, 0), fd.get_future_data(all_train_x, train_seq, 1, 0)], 'train_y': all_train_y, 'train_seq': train_seq, 'test_x': np.c_[all_test_x, fd.get_past_data(all_test_x, rows[:, 0], 1, 0), fd.get_past_data(all_test_x, rows[:, 0], 2, 0), fd.get_future_data(all_test_x, rows[:, 0], 1, 0)] } # Define prediction function
all_test_x = fd.load_submissions(files_txt) train_seq = fd.get_clean_sequences([1, 2, 3, 4, 5, 6, 7, 8, 9, 10]) rows, tmp = fd.load_test(['ds_pir_v0']) dataset = { 'train_x': all_train_x, 'train_y': all_train_y, 'train_seq': train_seq, 'test_x': all_test_x } # Add past data dataset = { 'train_x': np.c_[all_train_x, fd.get_past_data(all_train_x, train_seq, 1, -9999, 200), fd.get_past_data(all_train_x, train_seq, 2, -9999, 200), fd.get_future_data(all_train_x, train_seq, 1, -9999, 200)], 'train_y': all_train_y, 'train_seq': train_seq, 'test_x': np.c_[all_test_x, fd.get_past_data(all_test_x, rows[:, 0], 1, -9999), fd.get_past_data(all_test_x, rows[:, 0], 2, -9999), fd.get_future_data(all_test_x, rows[:, 0], 1, -9999)] } # Define prediction function
all_train_x, all_train_y, train_seq = fd.load_sequences(sequence_train, data_source) rows, all_test_x = fd.load_test(data_source) # Preprocess the whole data prepwd_params = {'remove_nan_targets':True, 'missing':0, 'float32':True, 'scale':True} all_train_x, all_train_y, train_seq, rows, all_test_x = fd.whole_preprocess(all_train_x, all_train_y, train_seq, rows, all_test_x, params=prepwd_params) # Add preprocessed data) all_train_x = np.concatenate((all_train_x, pd.read_csv('../predict_location/pl_L1_ET_vA1_valid.csv').values), axis=1) all_test_x = np.concatenate((all_test_x, pd.read_csv('../predict_location/pl_L1_ET_vA1_submission.csv').values[:, 3:]), axis=1) dataset = {'train_x':all_train_x, 'train_y':all_train_y, 'train_seq':train_seq, 'test_x':all_test_x} # Add past data dataset = {'train_x':np.c_[all_train_x, fd.get_past_data(all_train_x, train_seq, 1, 0), fd.get_past_data(all_train_x, train_seq, 2, 0), fd.get_future_data(all_train_x, train_seq, 1, 0)], 'train_y':all_train_y, 'train_seq':train_seq, 'test_x':np.c_[all_test_x, fd.get_past_data(all_test_x, rows[:,0], 1, 0), fd.get_past_data(all_test_x, rows[:,0], 2, 0), fd.get_future_data(all_test_x, rows[:,0], 1, 0)]} # Define prediction function def predict_model(train_x, train_y, test_x, test_y=None, class_weights=None, random_state=0): # Learn the KNN model params = {'layers': [[1000, 0.90], [100, 0.60]], 'loss': 'categorical_crossentropy'} model = fp.PMC_NeuralNetwork_T1(train_x.shape[1], train_y.shape[1], params, bags=1) model.fit(train_x, train_y, test_x, test_y, batch_size=64, nb_epoch=20)