processing = Processing_DB_Files()
project = Project()
#tuple from MPL
t_aux = []
for i in range(0,500):
t_aux.append(500)
t = tuple(t_aux)
####
classifiers = {"MPL": MLPClassifier(random_state=1, solver="adam", activation="relu", max_iter=100000, alpha=1e-5, hidden_layer_sizes=t), "Extratrees": ExtraTreesClassifier(n_estimators = 1000, random_state=1), "Knn":KNeighborsClassifier(n_neighbors=5), "Naive Bayes":GaussianNB(), "RandomForest":RandomForestClassifier(n_estimators = 1000, random_state=1), "Decision Tree":tree.DecisionTreeClassifier(random_state=1), "SVM":svm.SVC(probability=True, random_state=1)}
persons = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
get_accuracy = Get_Accuracy()
balance_data = BalanceData()
threshold_balance_data = 40
#Select the best classifier
accuracy_mean = pd.DataFrame(columns=["Classifier", "Accuracy"])
project.log("=====================ARCMA_SELECT_BEST_ALGORITHM=====================", file="arcma_best_algorithm.log")
for c in classifiers:
print(c)
person_accuracies = []
person_f_score = []
person_precision = []
person_recall = []
times_to_predict = []
for p in persons:
s = save()
try:
relevant_features = s.load_var("arcma_relevant_features_best_window{}relevant_features_{}.pkl".format(slash, p))
y = s.load_var("arcma_relevant_features_best_window{}y_{}.pkl".format(slash, p))
y = pd.DataFrame(y, columns=[arcma.label_tag])
except:
print("file from person {} not found!".format(p))
extra_trees = ExtraTreesClassifier(n_estimators=1000, random_state=0)
get_accuracy = Get_Accuracy()
balance_data = BalanceData()
threshold_balance_data = 40
#===LOAD FEATURES===#
#Interate threshold to find de best value#
persons = [
"f1", "m1", "m2", "f2", "m3", "f3", "m4", "m5", "m6", "m7", "f4", "m8",
"m9", "f5", "m10", "m11"
]
accuracy_by_person = pd.DataFrame()
threshold = 0.65
project.log(
"=========== HMP StratifiedKFold Accuracy, Thresold = {}===========".
format(threshold),
file="hmp_log_final_accuracy.log")
for p in persons:
s = save()
relevant_features = s.load_var(
"hmp_relevant_features_best_window{}relevant_features_{}.pkl".format(
slash, p))
y = s.load_var("hmp_relevant_features_best_window{}y_{}.pkl".format(
slash, p))
y = pd.DataFrame(y, columns=[hmp.label_tag])
skf = StratifiedKFold(n_splits=10, random_state=None, shuffle=False)
accuracy = {}
accuracies = []
balanced_data = balance_data.balance_data(relevant_features, y,
threshold_balance_data)
hmp = HMP_Model()
processing = Processing_DB_Files()
project = Project()
extra_trees = ExtraTreesClassifier(n_estimators=10000, random_state=0)
base_classification = Base_Classification(hmp, extra_trees)
#===LOAD FEATURES===#
#Interate threshold to find de best value#
s = save()
person_list = ["f1", "m1", "m2"]
accuracy_threshould_list = []
data = {}
threshold = 0.35
project.log(
"=========== HMP Outlier Accuracy, Thresold = {}===========".format(
threshold))
for p in person_list:
project.log("===========Person {}===========".format(p))
data = s.load_var("hmp_relevant_features{}relevant_features_{}.pkl".format(
slash, p))
y = s.load_var("hmp_relevant_features{}y_{}.pkl".format(slash, p))
y = pd.DataFrame(y, columns=[hmp.label_tag])
print("------------------------------------")
print("Person: {}".format(p))
print("------------------------------------")
return_accuracy = base_classification.get_accuracy.stratified_kfold_accuracy_outlier(
data, y, extra_trees, threshold, p)
project.log(str(return_accuracy), file="hmp_log.log")
#===INITIALIZATION===#
Debug.DEBUG = 0
arcma = ARCMA_Model()
processing = Processing_DB_Files()
project = Project()
persons = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
get_accuracy = Get_Accuracy()
balance_data = BalanceData()
threshold_balance_data = 40
#Select de best windows
t = time.time()
best_model = ExtraTreesClassifier(n_estimators = 1000, random_state=0)
w_accuracies = pd.DataFrame(columns=["window", "accurary"])
p = 15 # pessoa com mais registros
project.log("=====================ARCMA_SELECT_BEST_WINDOWS=====================", file="arcma_log_best_window.log")
for w in range(20,110,10):
print("Load data with window len = {}".format(w))
data = arcma.load_training_data_by_people(p)
print("Slicing Window....")
data_tsfresh, y = arcma.slice_by_window_tsfresh(data, w)
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, column_id='id', column_sort='time')
y = pd.DataFrame(y, columns=[arcma.label_tag])
balanced_data = balance_data.balance_data(relevant_features, y, threshold_balance_data)
column_id='id',
column_sort='time')
X_train, X_test, y_train, y_test = train_test_split(relevant_features,
y2,
test_size=0.2,
random_state=42)
extra_trees = ExtraTreesClassifier(n_estimators=10000,
max_depth=1000,
random_state=0)
extra_trees.fit(X_train, y_train)
start_time = time.time()
pred = extra_trees.predict(X_test)
end_time = time.time()
accuracy = accuracy_score(y_test, pred)
project.log("Accuracy to all ts_features ({}): {} - Time: {} seconds.".format(
len(relevant_features.columns), accuracy,
(end_time - start_time) / len(y_test)))
time.sleep(10)
del X_train, X_test, y_train, y_test, extra_trees, start_time, pred, accuracy
time.sleep(10)
# 2º - ACURÁCIA COM OS 10% MAIS RELEVANTES ts_features
ts_extratree_features_importance = pd.DataFrame(
extra_trees.feature_importances_,
index=X_train.columns,
columns=['importance']).sort_values('importance', ascending=False)
len_features = len(ts_extratree_features_importance)
best_features = ts_extratree_features_importance.index[0:int((len_features /
10) - 1)]
ts_final_features = relevant_features[best_features]
Debug.DEBUG = 0
arcma = ARCMA_Model()
processing = Processing_DB_Files()
project = Project()
extra_trees = ExtraTreesClassifier(n_estimators = 1000, random_state=0)
base_classification = Base_Classification(arcma, extra_trees)
balance_data = BalanceData()
threshold_balance_data = 40
#Interate threshold to find de best value#
s = save()
person_list = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
accuracy_threshould_temp_aux = pd.DataFrame(columns=["accuracy","discarted", "len_activity", "threshold"])
accuracy_mean = pd.DataFrame(columns=["accuracy","discarted", "len_activity", "threshold"])
project.log("====================ARCMA BEST THRESHOLD============================", file="arcma_log_best_threshold.log")
for t in np.arange(0.05, 1, 0.05):
accuracy_threshould_temp_aux = pd.DataFrame(columns=["accuracy","discarted", "len_activity"])
for p in person_list:
relevant_features = s.load_var("arcma_relevant_features_best_window{}relevant_features_{}.pkl".format(slash, p))
y = s.load_var("arcma_relevant_features_best_window{}y_{}.pkl".format(slash, p))
y = pd.DataFrame(y, columns=[arcma.label_tag])
balanced_data = balance_data.balance_data(relevant_features, y, threshold_balance_data)
if isinstance(balanced_data, tuple):
x_train, x_test, y_train, y_test = train_test_split(balanced_data[0], balanced_data[1], test_size=0.2, random_state=42)
data = {}
data[p] = {}
data[p]["training"] = {}
data[p]["training"]["training_features"] = x_train
data[p]["training"]["training_labels"] = y_train
#===INITIALIZATION===#
Debug.DEBUG = 0
umafall = UMAFALL_Model()
processing = Processing_DB_Files()
project = Project()
extra_trees = ExtraTreesClassifier(n_estimators = 10000, random_state=0)
base_classification = Base_Classification(umafall, extra_trees)
#===LOAD FEATURES===#
#Interate threshold to find de best value#
s = save()
person_list = [14,15, 16, 17]
accuracy_threshould_list = []
data = {}
threshold = 0.65
project.log("=========== UMAFALL Outlier Accuracy, Thresold = {}===========".format(threshold), file="umafall_log.log")
for p in person_list:
project.log("===========Person {}===========".format(p), file="umafall_log.log")
data = s.load_var("umafall_relevant_features{}relevant_features_{}.pkl".format(slash, p))
y = s.load_var("umafall_relevant_features{}y_{}.pkl".format(slash, p))
y = pd.DataFrame(y, columns=[umafall.label_tag])
print("------------------------------------")
print("Person: {}".format(p))
print("------------------------------------")
return_accuracy = base_classification.get_accuracy.stratified_kfold_accuracy_outlier(data, y, extra_trees, threshold, p, column_test=list(data.columns.values)[0])
project.log(str(return_accuracy), file="umafall_log.log")
project = Project()
extra_trees = ExtraTreesClassifier(n_estimators=10000, random_state=0)
base_classification = Base_Classification(arcma, extra_trees)
balance_data = BalanceData()
threshold_balance_data = 40
#===LOAD FEATURES===#
#Interate threshold to find de best value#
s = save()
person_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
accuracy_threshould_list = []
data = {}
threshold = 0.60
project.log(
"=========== ARCMA Outlier Accuracy, Thresold = {}===========".format(
threshold),
file="arcma_log_outlier_accuracy.log")
for p in person_list:
project.log("===========Person {}===========".format(p),
file="arcma_log_outlier_accuracy.log")
data = s.load_var(
"arcma_relevant_features_best_window{}relevant_features_{}.pkl".format(
slash, p))
y = s.load_var("arcma_relevant_features_best_window{}y_{}.pkl".format(
slash, p))
y = pd.DataFrame(y, columns=[arcma.label_tag])
print("------------------------------------")
print("Person: {}".format(p))
print("------------------------------------")
#===INITIALIZATION===#
Debug.DEBUG = 0
umafall = UMAFALL_Model()
processing = Processing_DB_Files()
project = Project()
persons = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
get_accuracy = Get_Accuracy()
balance_data = BalanceData()
threshold_balance_data = 40
#Select de best windows
t = time.time()
best_model = ExtraTreesClassifier(n_estimators=1000, random_state=0)
w_accuracies = pd.DataFrame(columns=["window", "accurary"])
p = 1 # pessoa com mais registros
project.log(
"=====================UMAFALL_SELECT_BEST_WINDOWS=====================",
file="umafall_log_best_window.log")
for w in range(10, 110, 10):
print("Load data with window len = {}".format(w))
data = umafall.load_training_data_by_people(
p, additional_where="and sensor=2 and body=2")
print("Slicing Window....")
data_tsfresh, y = umafall.slice_by_window_tsfresh(data, w)
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,
person_accuracies.append(accuracy)
times_to_predict.append(spent_time)
person_precision.append(precision)
person_recall.append(recall)
person_f_score.append(f_score)
person_discarteds.append(discarteds)
person_len_activities.append(len_activity)
except Exception as e:
print(e)
out_aux = pd.DataFrame({"Classifier":[type(classifiers[c]).__name__], "Threshold": t, "Accuracy":[st.mean(person_accuracies)], "Precision":[st.mean(person_precision)], "Recall":[st.mean(person_recall)], "F-Score":[st.mean(person_f_score)], "Time":[st.mean(times_to_predict)], "Discarteds":[st.mean(person_discarteds)], "Len Activities":[st.mean(person_len_activities)]})
accuracy_mean = pd.concat([accuracy_mean, out_aux])
accuracy_mean.to_csv(s.path+"new_results{}{}_best_threshold_balanced_data_window{}.csv".format(slash, model['model_name'], model['window']), sep='\t', encoding='utf-8')
'''
#Interate threshold to find de best value#
s = save()
person_list = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
accuracy_threshould_temp_aux = pd.DataFrame(columns=["accuracy","discarted", "len_activity", "threshold"])
accuracy_mean = pd.DataFrame(columns=["accuracy","discarted", "len_activity", "threshold"])
project.log("====================ARCMA BEST THRESHOLD============================", file="arcma_log_best_threshold.log")
for t in np.arange(0.05, 1, 0.05):
accuracy_threshould_temp_aux = pd.DataFrame(columns=["accuracy","discarted", "len_activity"])
for p in person_list:
relevant_features = s.load_var("arcma_relevant_features_best_window{}relevant_features_{}.pkl".format(slash, p))
y = s.load_var("arcma_relevant_features_best_window{}y_{}.pkl".format(slash, p))
y = pd.DataFrame(y, columns=[arcma.label_tag])
balanced_data = balance_data.balance_data(relevant_features, y, threshold_balance_data)