def train_random_forest(x_train, x_test, y_train, y_test, run, scenario):
"""
Função de treino do classificador Naive Bayes que retorna uma instancia treinada
:param x_train: Dados de treinamento
:param y_train: Classes dos dados de treinamento
:return: Classificador treinado
"""
clf = RandomForestClassifier(n_estimators=100, criterion='entropy')
clf = clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)
test_results = pd.DataFrame(data=[],
index=x_test.index.values.tolist())
test_results['pred_like'] = y_pred
test_results['original_like'] = y_test
positive_pred = test_results[test_results['pred_like'] == True]
candidate_songs = x_test.loc[positive_pred.index.values.tolist()]
if len(candidate_songs) == 0:
return pd.DataFrame()
# logging.info("User has no candidate songs!")
user_set = pd.concat([x_train, candidate_songs])
cos = pd.DataFrame(data=np.matrix(cosine_similarity(X=user_set)),
columns=user_set.index.tolist(),
index=user_set.index.tolist())
user_list = UserAverageController.start_ranking(
similarity_data_df=cos,
user_model_ids=x_train.index.values.tolist(),
song_model_ids=candidate_songs.index.values.tolist())
user_results_df = pd.concat([user_list, positive_pred],
axis=1,
sort=False)
return MachineAlgorithms.rank_evaluation(
run=run,
scenario=scenario,
algorithm="RF",
relevance_array=user_results_df['original_like'].values.tolist())
def train_linear_regressor(x_train, x_test, y_train, y_test, run,
scenario):
"""
Função de treino do classificador Naive Bayes que retorna uma instancia treinada
:param x_train: Dados de treinamento
:param y_train: Classes dos dados de treinamento
:return: Classificador treinado
"""
clf = LinearRegression()
clf = clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)
test_results = pd.DataFrame(data=[],
index=x_test.index.values.tolist())
test_results['pred_like'] = y_pred
test_results['original_like'] = y_test
positive_pred = test_results[test_results['pred_like'] == True]
candidate_songs = x_test.loc[positive_pred.index.values.tolist()]
if len(candidate_songs) == 0:
return pd.DataFrame()
# logging.info("User has no candidate songs!")
user_set = pd.concat([x_train, candidate_songs])
cos = pd.DataFrame(data=np.matrix(cosine_similarity(X=user_set)),
columns=user_set.index.tolist(),
index=user_set.index.tolist())
user_list = UserAverageController.start_ranking(
similarity_data_df=cos,
user_model_ids=x_train.index.values.tolist(),
song_model_ids=candidate_songs.index.values.tolist())
user_results_df = pd.concat([user_list, positive_pred],
axis=1,
sort=False)
map_value = MAPController.get_ap_from_list(
user_results_df['original_like'].tolist())
mrr_value = MRRController.get_rr_from_list(
user_results_df['original_like'].tolist())
output = pd.concat([
pd.DataFrame(data=[[run, scenario, 'LR', 'map', map_value]],
columns=GlobalVariable.results_column_name),
pd.DataFrame(data=[[run, scenario, 'LR', 'mrr', mrr_value]],
columns=GlobalVariable.results_column_name)
])
return output
def user_average(x_train_data, x_test_data, y_test_label, run, scenario):
test_df = pd.DataFrame(data=[],
index=x_test_data.index.values.tolist())
test_df['original_like'] = y_test_label
user_set = pd.concat([x_train_data, x_test_data])
cos = pd.DataFrame(data=np.matrix(cosine_similarity(X=user_set)),
columns=user_set.index.tolist(),
index=user_set.index.tolist())
user_list = UserAverageController.start_ranking(
similarity_data_df=cos,
user_model_ids=x_train_data.index.values.tolist(),
song_model_ids=x_test_data.index.values.tolist())
user_results_df = pd.concat([user_list, test_df],
axis=1,
sort=False,
join='inner')
return MachineAlgorithms.rank_evaluation(
run=run,
scenario=scenario,
algorithm="recModel",
relevance_array=user_results_df['original_like'].values.tolist())
def train_perceptron(x_train, x_test, y_train, y_test, run, scenario):
"""
Função de treino do classificador Perceptron que retorna uma instancia treinada
As constantes estão no arquivo de variaveis de sistema
:param x_train: Dados de treinamento
:param y_train: Classes dos dados de treinamento
:return: Classificador treinado
"""
clf = Perceptron(max_iter=1000,
tol=1e-3,
n_jobs=GlobalVariable.processor_number)
clf.fit(x_train, y_train)
y_pred = clf.predict(x_test)
test_results = pd.DataFrame(data=[],
index=x_test.index.values.tolist())
test_results['pred_like'] = y_pred
test_results['original_like'] = y_test
positive_pred = test_results[test_results['pred_like'] == True]
candidate_songs = x_test.loc[positive_pred.index.values.tolist()]
if len(candidate_songs) == 0:
return pd.DataFrame()
# logging.info("User has no candidate songs!")
user_set = pd.concat([x_train, candidate_songs])
cos = pd.DataFrame(data=np.matrix(cosine_similarity(X=user_set)),
columns=user_set.index.tolist(),
index=user_set.index.tolist())
user_list = UserAverageController.start_ranking(
similarity_data_df=cos,
user_model_ids=x_train.index.values.tolist(),
song_model_ids=candidate_songs.index.values.tolist())
user_results_df = pd.concat([user_list, positive_pred],
axis=1,
sort=False)
return MachineAlgorithms.rank_evaluation(
run=run,
scenario=scenario,
algorithm="PER",
relevance_array=user_results_df['original_like'].values.tolist())