def __init__(self, model: Model, clf):
self.model = model
self.get_accuracy = Get_Accuracy()
self.processing = Processing_DB_Files()
self.clf = clf
super().__init__()
# -*- coding: utf-8 -*-
# IMPORTS #
from utils.debug import Debug
from models.arcma_model import ARCMA_Model
from tsfresh import extract_relevant_features
from pre_processing.processing_db_files import Processing_DB_Files
from utils.project import Project, slash
from scripts.save_workspace import save
#===INITIALIZATION===#
Debug.DEBUG = 0
arcma = ARCMA_Model()
processing = Processing_DB_Files()
project = Project()
s = save()
#window = 26 # Janela Fixa
window = 50 # Melhor Janela
persons = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
for p in persons:
data = arcma.load_training_data_by_people(p)
print("Slicing Window....")
data_tsfresh, y = arcma.slice_by_window_tsfresh(data, window)
y.index += 1
del data_tsfresh["activity"]
classes_counts = y.value_counts()
if len(classes_counts) > 1:
relevant_features = extract_relevant_features(data_tsfresh,
y,
def outlier_prepare(self, model:Model, people:str, windows_size:int, activity_outlier:str):
processing = Processing_DB_Files()
model.load_training_data_by_window_by_people(people, windows_size)
training, training_labels, test, test_labels = processing.calculating_features(model)
return self.generate_outliers(training, training_labels, test, test_labels, activity_outlier)
from tsfresh import extract_features
from tsfresh import extract_relevant_features
from tsfresh.utilities.dataframe_functions import impute
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from pre_processing.processing_db_files import Processing_DB_Files
import itertools
from sklearn.metrics import accuracy_score
from utils.project import Project
import time
#===INITIALIZATION===#
Debug.DEBUG = 0
hmp = HMP_Model()
processing = Processing_DB_Files()
project = Project()
extra_trees = ExtraTreesClassifier(n_estimators=1000,
max_depth=1000,
random_state=0) #Good performer
base_classification = Base_Classification(hmp, extra_trees)
_, _, _ = base_classification.predict_outliers_for_list_people_with_proba(
36, ["f1", "m1", "m2"], "eat_soup", 0.55, remove_outliers=0.05)
#===Extract TsFresh Features===#
dataframe_1 = hmp.data_with_window["f1"]["training"]
dataframe_2 = pd.DataFrame()
labels = []
id = 1
for d in dataframe_1:
if len(np.unique(d[hmp.label_tag])) < 2:
class Base_Classification(object):
def __init__(self, model: Model, clf):
self.model = model
self.get_accuracy = Get_Accuracy()
self.processing = Processing_DB_Files()
self.clf = clf
super().__init__()
#Finding the best window to improve the accuracy
def find_best_window(self, interval, list_people):
accuracies = []
for w in interval:
Debug.print_debug("WINDOWS SIZE: {}".format(w))
data_all_people = self.model.load_training_data_from_list_people(
w, list_people)
features_all_people = self.processing.calculating_features_to_each_person(
data_all_people, self.model)
accuracies_dict = self.get_accuracy.simple_accuracy_mean_to_each_person(
features_all_people, self.model, self.clf)
accuracy_mean = mean(accuracies_dict.values())
accuracies.append({'window': w, 'accuracy': accuracy_mean})
return accuracies
def predict_for_all_people_with_proba(self, window, threshold):
data_all_people = self.model.load_training_data_from_all_people(window)
features_all_people = self.processing.calculating_features_to_each_person(
data_all_people, self.model)
accuracies_proba = self.get_accuracy.simple_accuracy_mean_to_each_person_with_proba(
features_all_people, self.model, self.clf, threshold)
accuracies = self.get_accuracy.simple_accuracy_mean_to_each_person(
features_all_people, self.model, self.clf)
return accuracies, accuracies_proba
def predict_for_list_people_with_proba(self, window, list_people,
threshold):
data_list_people = self.model.load_training_data_from_list_people(
window, list_people)
features_all_people = self.processing.calculating_features_to_each_person(
data_list_people, self.model)
accuracies_proba = self.get_accuracy.simple_accuracy_mean_to_each_person_with_proba(
features_all_people, self.model, self.clf, threshold)
accuracies = self.get_accuracy.simple_accuracy_mean_to_each_person(
features_all_people, self.model, self.clf)
return accuracies, accuracies_proba
def predict_outliers_for_list_people_with_proba(self,
window,
list_people,
activity,
threshold,
remove_outliers=0):
data_list_people = self.model.load_training_data_from_list_people(
window, list_people, remove_outliers)
features_all_people = self.processing.calculating_features_to_each_person(
data_list_people, self.model)
#return self.get_accuracy.simple_accuracy_outlier_activity(features_all_people, self.model, self.clf, activity,threshold)
return self.get_accuracy.get_outliers_confused_with_activities(
features_all_people, self.model, self.clf, threshold)
training, training_labels, test, test_labels, outlier, outlier_labels = self.get_accuracy.simple_accuracy_outlier_activity(
features_all_people, self.model, self.clf, activity, threshold)
# -*- coding: utf-8 -*-
from utils.debug import Debug
from models.hmp_model import HMP_Model
from pre_processing.processing_db_files import Processing_DB_Files
from outlier.outlier_commons import Outlier_Commons
# INITIALIZATION #
Debug.DEBUG = 1
hmp = HMP_Model()
outlier = Outlier_Commons()
# PROCESSING #
processing = Processing_DB_Files()
hmp.load_training_data_by_window_by_people('f1', 50)
training, training_labels, test, test_labels = processing.calculating_features(
hmp)
training_outlier, training_labels_outlier, test_outlier, test_labels_outlier, \
outlier, outlier_labels = outlier.generate_outliers(training, training_labels, test, test_labels, "drink_glass")