def knn_by_minute(data, features, prefix, name, minute):
features = get_features(features, prefix, minute)
target = 'result'
train_x, train_y, test_x, test_y = utils.split_train_test(
data, features, target)
dirname = PATH_JOBS_MODELS + name + "/"
model_file = dirname + "2_" + str(minute) + ".joblib"
with open(model_file, "rb") as f:
model = pickle.load(f)
predict_knn(minute, model, name, test_x, test_y, train_x, train_y)
def gnb_by_minute(data, features, prefix, name, minute):
features = get_features(features, prefix, minute)
target = 'result'
train_x, train_y, test_x, test_y = utils.split_train_test(
data, features, target)
dirname = PATH_JOBS_MODELS + name + "/"
model_file = dirname + str(minute) + ".joblib"
with open(model_file, "rb") as f:
model = pickle.load(f)
pd.options.mode.chained_assignment = None # default='warn'
predict(minute, model, name, test_x, test_y, train_x, train_y)
def gnb_by_minute(data, features, prefix, name, minute):
features = get_features(features, prefix, minute)
target = 'result'
train_x, train_y, test_x, test_y = utils.split_train_test(
data, features, target)
model = GaussianNB()
model.fit(train_x, train_y)
dirname = PATH_JOBS_MODELS + name + "/"
if not os.path.exists(dirname):
os.mkdir(dirname)
model_file = dirname + str(minute) + ".joblib"
with open(model_file, "wb") as f:
pickle.dump(model, f)
evaluate = True
if evaluate:
testing.predict(minute, model, name, test_x, test_y, train_x, train_y)
def xgb_by_minute(data, features, prefix, name, minute, odds):
features = get_features(features, prefix, minute)
target = 'result'
if odds:
features.append('odds_p_avg_home')
features.append('odds_p_avg_draw')
features.append('odds_p_avg_away')
features.append('odds_p_std_home')
features.append('odds_p_std_draw')
features.append('odds_p_std_away')
train_x, train_y, test_x, test_y = utils.split_train_test(
data, features, target, odds)
dirname = PATH_JOBS_MODELS + name + "/"
model_file = dirname + "1_" + str(minute) + ".joblib"
with open(model_file, "rb") as f:
model = pickle.load(f)
predict_knn(minute, model, name, test_x, test_y, train_x, train_y)
def mlp_by_minute(data, features, prefix, name, config):
cnn_encoder_kernel_sizes, cnn_encoder_layer_sizes, minute, optim_batch_size, optim_learning_rate, \
optim_num_epochs, method, dropout = utils.get_config(config)
features = utils.get_features(features, prefix, minute)
target = 'result'
train_x, train_y, test_x, test_y = utils.split_train_test(
data, features, target)
def get_target(row):
if row == 'H':
return 0
elif row == 'D':
return 1
elif row == 'A':
return 2
train_y = train_y.apply(lambda row: get_target(row))
train_y_cat = to_categorical(train_y)
test_y = test_y.apply(lambda row: get_target(row))
test_y_cat = to_categorical(test_y)
model_dir = PATH_JOBS_MODELS + 'mlp_stats' + "/"
model_label = config['minute'] + "_" + config['method'] + "_" + \
config['optim_num_epochs'] + "_" + \
config['optim_batch_size'] + "_" + \
config['optim_learning_rate'] + "_" + \
config['cnn_encoder_layer_sizes'] + "_" + config['optim:dropout_rate']
model_file_json = model_dir + model_label + ".json"
model_file_h5 = model_dir + model_label + ".h5"
# load json and create model
json_file = open(model_file_json, 'r')
loaded_model_json = json_file.read()
json_file.close()
loaded_model = model_from_json(loaded_model_json)
# load weights into new model
loaded_model.load_weights(model_file_h5)
optimizer = Adam(lr=optim_learning_rate)
loaded_model.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy', rps_loss])
eval_train = loaded_model.evaluate(train_x, train_y_cat)
eval_test = loaded_model.evaluate(test_x, test_y_cat)
print(eval_train)
print(eval_test)
met = pd.DataFrame()
met["acc_train"] = [eval_train[1]]
met["rps_train"] = [eval_train[2]]
met["acc_test"] = [eval_test[1]]
met["rps_test"] = [eval_test[2]]
dirname = PATH_JOBS_RESULTS + name + "/"
if not os.path.exists(dirname):
os.mkdir(dirname)
results_file = dirname + str(minute) + ".csv"
met.to_csv(results_file)
print(met)