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)
# load model
model_name = 'bhmm'
ctrl.load_model(MODEL_FILE_PATH, model_name)
ctrl.create_model_agnostics(model_name)
hm = ctrl.get_model(model_name)._obs_lbl_hashmap
file_path = MODEL_FOLDER_PATH + '/' + MODEL_NAME + '.lime_sleeping.png'
file_paths = [file_path]
labels_to_explain=['sleeping']
import numpy as np
# typical for activation pattern for sleeping
raw_feature_sleeping = np.array(
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], dtype=bool)
ch_feature_sleeping = np.array(
[0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=bool)
lf_feature_sleeping = np.array(
[0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype=bool)
data = ctrl._dataset # type: _Dataset
ctrl.plot_and_save_explanation(
model_name,
ch_feature_sleeping,
labels_to_explain,
file_paths
)
def main():
ctrl = Controller()
ctrl.load_dataset_from_file(DATASET_FILE_PATH)
# load model
model_name = 'bhmm'
ctrl.load_model(MODEL_FILE_PATH, model_name)
ctrl.create_model_agnostics(model_name)
file_path = MODEL_FOLDER_PATH + '/' + MODEL_NAME + '.feature_importance.png'
ctrl.save_plot_feature_importance(model_name, file_path)
def main():
ctrl = Controller()
ctrl.load_dataset_from_file(DATASET_FILE_PATH)
print()
# load model
ctrl.load_model(MODEL_FILE_PATH, 'bhmm')
ctrl.load_model(MODEL_FILE_PATH2, 'bhsmm')
state_sel = ['dental_care', 'outside_activity', 'learning']
#state_sel = ['dental_care', 'enter_home', 'kitchen_social_activity']
img_file_path = MODEL_FOLDER_PATH + '/' + 'comparision.png'
ctrl.compare_dur_dists_and_save('bhmm', 'bhsmm', state_sel, img_file_path)
class TestPendigitsModel(unittest.TestCase):
def setUp(self):
self.ctrl = Controller()
self.ctrl.load_dataset(Dataset.PENDIGITS)
pd_model = ModelPendigits(self.ctrl, "test")
self.ctrl.register_model(pd_model)
self.dset = self.ctrl._dataset # type: DatasetPendigits
self.pd_model = self.ctrl._model
def test_init_hmms(self):
self.ctrl.init_model_on_dataset()
def test_save_hmms(self):
self.ctrl.init_model_on_dataset()
self.ctrl.save_model()
def test_load_hmms(self):
self.ctrl.load_model()
# attention manually refresh reference
self.pd_model = self.ctrl._model
def test_train_hmms(self):
self.ctrl.init_model_on_dataset()
self.ctrl.train_model()
def test_train_n_save_hmms(self):
self.ctrl.init_model_on_dataset()
self.ctrl.train_model()
self.ctrl.save_model()
def test_pre_bench_hmms(self):
self.ctrl.load_model()
self.pd_model = self.ctrl._model
y_true, y_pred = self.pd_model.create_pred_act_seqs(self.dset)
def test_bench_hmms(self):
self.ctrl.load_model()
self.pd_model = self.ctrl._model
self.ctrl.register_benchmark()
rep = self.ctrl.create_report(conf_matrix=True,
accuracy=True,
precision=True,
recall=True,
f1=True)
print(rep)
def test_generate_obs(self):
self.ctrl.load_model()
self.pd_model = self.ctrl._model
self.ctrl.generate_observations()
def test_generate_obs_n_plot(self):
# numbers that were good in benchmark (desc):
# [0,8,4,9,3,7]
self.ctrl.load_model()
ds = self.ctrl._dataset # type: DatasetPendigits
pdmod = self.ctrl._model # type: ModelPendigits
num = 1
zero_start = [8, 1, 1]
one_start = [8, 2, 1, 1, 1]
start_seq = one_start
hmm = pdmod._model_dict[num]
hmm.set_format_full(True)
print(hmm)
pdmod.select_number(num)
seq = pdmod.generate_observations(start_seq)
ds.plot_obs_seq(seq, num)
def tearDown(self):
pass
class TestController(unittest.TestCase):
def setUp(self):
# set of observations
self.ctrl = Controller() # type: Controller
#self.hmm_model = ModelHMM(self.ctrl)
#self.hmm_model = ModelHMM_scaled(self.ctrl)
#self.hmm_model = ModelHMM_log(self.ctrl)
self.hmm_model = ModelHMM_log_scaled(self.ctrl)
def tearDown(self):
pass
def test_ctrl_get_bench_metrics(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
self.ctrl.register
# render
#dot = self.ctrl.render_model()
#dot.render('test.gv', view=True)
self.ctrl.train_model(True)
print(self.ctrl.get_bench_metrics())
def test_ctrl_train_seqs(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
self.ctrl.load_dataset()
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
# render
#dot = self.ctrl.render_model()
#dot.render('test.gv', view=True)
self.ctrl.train_model()
report = self.ctrl.create_report(conf_matrix=True,
accuracy=True,
precision=True,
recall=True,
f1=True)
print(self.ctrl._model)
print(report)
#self.ctrl.show_plot()
def test_ctrl_presentation(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
# render
#dot = self.ctrl.render_model()
#dot.render('test.gv', view=True)
self.ctrl.train_model(True)
report = self.ctrl.create_report(conf_matrix=True,
accuracy=True,
precision=True,
recall=True,
f1=True)
print(self.ctrl._model)
print(report)
#self.ctrl.show_plot()
# render
#dot = self.ctrl.render_model()
#dot.render('test.gv', view=True)
def test_generate_visualization(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
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_label_hm: ', hmm_model._state_lbl_hashmap)
print('state_label_rev_hm: ', hmm_model._state_lbl_rev_hashmap)
self.ctrl.save_visualization_to_file(
'/home/cmeier/code/tmp/visualization.png')
def test_bench_train_loss(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
self.ctrl.load_dataset()
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
self.ctrl.register_loss_file_path(
'/home/cmeier/code/tmp/kasteren/train_loss.log')
self.ctrl.train_model()
self.ctrl.save_model(
'/home/cmeier/code/tmp/kasteren/kasteren_model.joblib')
def test_bench_reports_conf_matrix(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
self.ctrl.train_model()
report = self.ctrl.create_report(conf_matrix=True)
print(report)
def test_bench_reports(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
self.ctrl.train_model()
report = self.ctrl.create_report(conf_matrix=True,
accuracy=True,
precision=True,
recall=True,
f1=True)
print(report)
#self.ctrl.show_plot()
def test_bench_q_fct(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
self.ctrl.register_benchmark()
print(self.ctrl._model)
# use dataset Kasteren and q_fct
self.ctrl.train_model(True)
report = self.ctrl.create_report()
print(self.ctrl._model)
print(report)
#self.ctrl.show_plot()
def test_om(self):
#plt.figure(figsize=(10,6))
#self.pom.plot()
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
dot = self.ctrl.render_model()
dot.render('test.gv', view=True)
def test_train_model(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
self.ctrl.load_dataset()
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
#print(hmm_model._hmm)
self.ctrl.train_model()
print('#' * 200)
print(hmm_model)
hmm_model._hmm.set_str_exp(True)
hmm_model._hmm.set_format_full(True)
print(hmm_model)
#print(hmm_model._hmm.verify_transition_matrix())
#print(hmm_model._hmm.verify_emission_matrix())
#self._bench._model.draw()
#self.ctrl.register_benchmark()
#report = self.ctrl.create_report(True, True, True, True, True)
#print(report)
#self._bench.show_plot()
def test_save_model(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.load_dataset(dk)
self.ctrl.register_model(hmm_model)
self.ctrl.init_model_on_dataset()
hmm_model._hmm.set_format_full(True)
self.ctrl.train_model()
self.ctrl.save_model()
def test_init_model(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
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_model(self):
hmm_model = self.hmm_model
dk = Dataset.KASTEREN
self.ctrl.set_dataset(dk)
self.ctrl.load_dataset()
self.ctrl.register_model(hmm_model)
self.ctrl.load_model()
print(self.ctrl._model)