def main(): ctrl = Controller() ctrl.load_dataset_from_file(DATASET_FILE_PATH) if MODE == MODE_TRAIN: from scripts.test_model import BHSMMTestModel from hassbrain_algorithm.models.hmm.bhmm_hp import BernoulliHMM_HandcraftedPriors from hassbrain_algorithm.models.tads import TADS if MODEL_CLASS == BHMM: hmm_model = BHMMTestModel(ctrl) elif MODEL_CLASS == BHSMM: hmm_model = BHSMMTestModel(ctrl) hmm_model.set_training_steps(50) elif MODEL_CLASS == BHMMPC: hmm_model = BernoulliHMM_HandcraftedPriors(ctrl) elif MODEL_CLASS == MCTADS: hmm_model = TADS(ctrl) else: raise ValueError ctrl.register_model(hmm_model, MODEL_NAME) # load domain knowledge if MODEL_CLASS == BHMMPC: path = '/home/cmeier/code/data/hassbrain/datasets/hass_chris_final/data/domain_knowledge.json' act_data, loc_data = load_domain_knowledge(path) ctrl.register_location_info(MODEL_NAME, loc_data) ctrl.register_activity_info(MODEL_NAME, act_data) # load model elif MODE == MODE_BENCH: ctrl.load_model(MODEL_FILE_PATH, MODEL_NAME) else: raise ValueError ctrl.register_benchmark(MODEL_NAME) ctrl.init_model_on_dataset(MODEL_NAME) if MODE == MODE_TRAIN: ctrl.register_loss_file_path(MD_LOSS_FILE_PATH, MODEL_NAME) ctrl.train_model(MODEL_NAME) ctrl.save_model(MODEL_FILE_PATH, MODEL_NAME) # bench the model reports = ctrl.bench_models() # save metrics ctrl.save_df_metrics_to_file(MODEL_NAME, MD_METRICS_FILE_PATH) ctrl.save_df_confusion(MODEL_NAME, MD_CONF_MAT_FILE_PATH) ctrl.save_df_act_dur_dists(MODEL_NAME, MD_ACT_DUR_DISTS_DF_FILE_PATH, DATA_ACT_DUR_DISTS_DF_FILE_PATH) ctrl.save_df_class_accs(MODEL_NAME, MD_CLASS_ACTS_FILE_PATH) # plots if MODE == MODE_TRAIN and MODEL_CLASS != MCTADS: ctrl.save_plot_trainloss(MODEL_NAME, MD_LOSS_IMG_FILE_PATH) ctrl.save_plot_inferred_states(MODEL_NAME, MD_INFST_IMG_FILE_PATH) ctrl.save_plot_act_dur_dists([MODEL_NAME], MD_ACT_DUR_DISTS_IMG_FILE_PATH)
def main(): ctrl = Controller() ctrl.load_dataset_from_file(DATASET_FILE_PATH) #hmm_model = BHMMTestModel(ctrl) #hmm_model = BHSMMTestModel(ctrl) from hassbrain_algorithm.models.hmm.bhmm_hp import BernoulliHMM_HandcraftedPriors hmm_model = BernoulliHMM_HandcraftedPriors(ctrl) ctrl.register_model(hmm_model, MODEL_NAME) # load domain knowledge path = '/home/cmeier/code/data/hassbrain/datasets/hass_chris_final/data/domain_knowledge.json' act_data, loc_data = load_domain_knowledge(path) ctrl.register_location_info(MODEL_NAME, loc_data) ctrl.register_activity_info(MODEL_NAME, act_data) # load model #ctrl.load_model(MODEL_FILE_PATH, MODEL_NAME) from scripts.test_model import BHSMMTestModel ctrl.register_benchmark(MODEL_NAME) ctrl.init_model_on_dataset(MODEL_NAME) ctrl.register_loss_file_path(MD_LOSS_FILE_PATH, MODEL_NAME) ctrl.train_model(MODEL_NAME) # bench the model reports = ctrl.bench_models() # save metrics ctrl.save_df_metrics_to_file(MODEL_NAME, MD_METRICS_FILE_PATH) ctrl.save_df_confusion(MODEL_NAME, MD_CONF_MAT_FILE_PATH) ctrl.save_df_act_dur_dists(MODEL_NAME, MD_ACT_DUR_DISTS_DF_FILE_PATH, DATA_ACT_DUR_DISTS_DF_FILE_PATH) # plots ctrl.save_plot_trainloss(MD_LOSS_IMG_FILE_PATH, MODEL_NAME) ctrl.plot_and_save_inferred_states(MD_INFST_IMG_FILE_PATH, MODEL_NAME) ctrl.save_plot_act_dur_dists(MODEL_NAME, MD_ACT_DUR_DISTS_IMG_FILE_PATH, DATA_ACT_DUR_DISTS_IMG_FILE_PATH)
class TestHomeassistantModelHMMLogScaled(unittest.TestCase): # Model part def setUp(self): # set of observations self.ctrl = Controller() self.ctrl.set_dataset(Dataset.HASS_TESTING) self.hass_obj = self.ctrl._dataset #type: DatasetHomeassistant self.hmm_model = PreConfHMM(self.ctrl) def tearDown(self): pass def test_load_custom_lists_modelHMM(self): custom_state_list = ['sleeping', 'cooking'] custom_obs_list = [ 'binary_sensor.motion_bed', 'binary_sensor.motion_mirror', 'binary_sensor.motion_pc', 'switch.test_switch_1', 'light.test_light' ] hmm_model = self.hmm_model self.ctrl.set_custom_state_list(custom_state_list) self.ctrl.set_custom_obs_list(custom_obs_list) self.ctrl.load_dataset() self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() hmm_model._hmm.set_format_full(True) print(self.ctrl._model) def test_load_modelHMM(self): self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() hmm_model._hmm.set_format_full(True) print(self.ctrl._model) print(self.hass_obj.get_obs_lbl_hashmap()) print(self.hass_obj.get_state_lbl_hashmap()) def test_train_modelHMM(self): self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() hmm_model._hmm.set_format_full(True) print(self.ctrl._model) self.ctrl.train_model() print(self.ctrl._model) def test_bench_modelHMM(self): self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() hmm_model._hmm.set_format_full(True) print(self.ctrl._model) self.ctrl.register_benchmark() self.ctrl.train_model() print(self.ctrl._model) report = self.ctrl.create_report( conf_matrix=True, accuracy=True, precision=True, recall=True, f1=True ) print(report) def test_classify(self): self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() self.ctrl.train_model() hmm_model._hmm.set_format_full(True) print(hmm_model) print('-'*10) obs_seq = [('binary_sensor.motion_bed', 0), ('binary_sensor.motion_mirror', 1), ('binary_sensor.motion_bed', 0)] pred_state = hmm_model.classify(obs_seq) print('#'*100) print(pred_state) def test_classify_multi(self): """ used to test for classification of multiple labels """ self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() self.ctrl.train_model() hmm_model._hmm.set_format_full(True) print(hmm_model) print('-'*10) obs_seq = [('binary_sensor.motion_bed', 0), ('binary_sensor.motion_mirror', 1)]#, ('binary_sensor.motion_bed', 0)] act_state_dict = hmm_model.classify_multi(obs_seq) print('#'*100) print(act_state_dict) #print(act_state_dict) #print(hmm_model.get_state_label_list()) def test_pred_next_obs_single(self): self.ctrl.load_dataset() hmm_model = self.hmm_model self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() self.ctrl.train_model() hmm_model._hmm.set_format_full(True) #print(hmm_model) print('#'*100) obs_seq = [('binary_sensor.motion_bed', 0), ('binary_sensor.motion_mirror', 1), ('binary_sensor.motion_bed', 0)] tupel = hmm_model.predict_next_obs(obs_seq) print(tupel) def test_pred_next_obs_multi(self): hmm_model = self.hmm_model self.ctrl.load_dataset() self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() self.ctrl.train_model() hmm_model._hmm.set_format_full(True) print(hmm_model) print('#'*100) obs_seq = [('binary_sensor.motion_bed', 0), ('binary_sensor.motion_mirror', 1), ('binary_sensor.motion_bed', 0)] #arr = hmm_model.predict_next_obs_arr(obs_seq) print(hmm_model._obs_lbl_hashmap) print(hmm_model._obs_lbl_rev_hashmap) res_dict = hmm_model.predict_prob_xnp1(obs_seq) print(hmm_model._obs_lbl_hashmap) print(hmm_model._obs_lbl_rev_hashmap) res_dict = hmm_model.predict_prob_xnp1(obs_seq) print(hmm_model._obs_lbl_hashmap) print(hmm_model._obs_lbl_rev_hashmap) print('#'*100) print(res_dict) def test_encode_loc_data(self): loc_data = [ { "name" : "loc1", "activities" : ['cooking'], "devices" : ['binary_sensor.motion_hallway', 'binary_sensor.motion_mirror'], }, {"name" : "loc2", "activities" : ['cooking', 'eating'], "devices" : [], }, {"name" : "loc3", "activities" : ['sleeping'], "devices" : ['binary_sensor.motion_bed'], }, ] hmm_model = self.hmm_model self.ctrl.load_dataset() self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() hmm_model._hmm.set_format_full(True) #print('state_hm: ', hmm_model._state_lbl_hashmap) #print('obs_hm: ', hmm_model._obs_lbl_hashmap) #print('raw_loc_data: \t' + str(loc_data)) enc_loc_data = hmm_model._encode_location_data(loc_data) #print('#'*100) #print('enc_loc_data: \t' + str(enc_loc_data)) def test_encode_act_data(self): act_data = [ {"name" : "cooking", "day_of_week" : 2, "start" : datetime.time.fromisoformat("06:15:00"), "end" : datetime.time.fromisoformat("08:45:00") }, {"name" : "eating", "day_of_week" : 1, "start" : datetime.time.fromisoformat("06:15:00"), "end" : datetime.time.fromisoformat("08:45:00") }, {"name" : "eating", "day_of_week" : 1, "start" : datetime.time.fromisoformat("08:46:00"), "end" : datetime.time.fromisoformat("10:00:00") }, ] hmm_model = self.hmm_model self.ctrl.load_dataset() self.ctrl.register_model(hmm_model) self.ctrl.init_model_on_dataset() print('raw_act_data: \t' + str(act_data)) print('state_hm: ', hmm_model._state_lbl_hashmap) print('obs_hm: ', hmm_model._obs_lbl_hashmap) print('#'*100) enc_act_data = hmm_model._encode_act_data(act_data) print('enc_act_data: \t' + str(enc_act_data)) def test_init(self): loc_data = [ { "name" : "loc1", "activities" : ['cooking'], "devices" : ['binary_sensor.motion_hallway', 'binary_sensor.motion_mirror'], }, {"name" : "loc2", "activities" : ['cooking', 'eating'], "devices" : [], }, {"name" : "loc3", "activities" : ['sleeping'], "devices" : ['binary_sensor.motion_bed'], }, ] act_data = [ {"name" : "sleeping", "day_of_week" : 0, "start" : datetime.time.fromisoformat("04:00:00"), "end" : datetime.time.fromisoformat("06:15:00") }, {"name" : "cooking", "day_of_week" : 0, "start" : datetime.time.fromisoformat("06:15:00"), "end" : datetime.time.fromisoformat("08:45:00") }, {"name" : "eating", "day_of_week" : 0, "start" : datetime.time.fromisoformat("08:46:00"), "end" : datetime.time.fromisoformat("10:00:00") }, {"name" : "sleeping", "day_of_week" : 0, "start" : datetime.time.fromisoformat("12:00:00"), "end" : datetime.time.fromisoformat("13:00:00") }, {"name" : "sleeping", "day_of_week" : 1, "start" : datetime.time.fromisoformat("02:00:00"), "end" : datetime.time.fromisoformat("06:30:00") }, {"name" : "cooking", "day_of_week" : 1, "start" : datetime.time.fromisoformat("12:00:00"), "end" : datetime.time.fromisoformat("13:00:00") }, {"name" : "cooking", "day_of_week" : 2, "start" : datetime.time.fromisoformat("19:00:00"), "end" : datetime.time.fromisoformat("00:00:00") }, {"name" : "cooking", "day_of_week" : 2, "start" : datetime.time.fromisoformat("23:00:00"), "end" : datetime.time.fromisoformat("00:00:00") }, {"name" : "sleeping", "day_of_week" : 2, "start" : datetime.time.fromisoformat("00:00:00"), "end" : datetime.time.fromisoformat("03:00:00") }, ] hmm_model = self.hmm_model self.ctrl.load_dataset() self.ctrl.register_model(hmm_model) self.ctrl.register_location_info(loc_data) print('raw_act_data: \t' + str(act_data)) print('state_hm: ', hmm_model._state_lbl_hashmap) #print('obs_hm: ', hmm_model._obs_lbl_hashmap) self.ctrl.register_activity_info(act_data) #print('#'*100) #enc_act_data = hmm_model._encode_act_data(act_data) #print('enc_act_data: \t' + str(enc_act_data)) self.ctrl.init_model_on_dataset() hmm = hmm_model._hmm hmm.set_format_full(True) self.assertAlmostEqual(1.0, Probs.prob_to_norm(hmm._pi.sum()), 6) self.assertTrue(hmm.verify_emission_matrix()) self.assertTrue(hmm.verify_transition_matrix())