def get_score_for_clf(self, clf, df_test, feature_name, pca=None):
x_test = df_test[self.feature_columns_dict[feature_name]]
x_test_pca = x_test[[col for col in x_test.columns if 'gdf' in col]]
x_test = x_test[[col for col in x_test.columns if 'gdf' not in col]]
if pca:
x_test_pca = pca.transform(x_test_pca)
for n in range(pca.n_components):
x_test['pca_{}'.format(n)] = x_test_pca[:, n]
y_test = df_test['mid_price_indicator'].values
return model.test_model(clf, x_test, y_test)
def get_score_for_clf(self, clf, df_test, feature_name, pca=None):
x_test = df_test[self.feature_columns_dict[feature_name]]
if pca:
x_test = pca.transform(x_test)
y_test = df_test['mid_price_indicator'].values
return model.test_model(clf, x_test, y_test)
def train_lstm(self,
clf,
feature_name='',
should_validate=True,
method=None,
fit_kwargs=None,
compile_kwargs=None,
n_steps=None,
plot_name=None,
class_weight=None,
should_return_model=False):
logger.info('Training %s r=%s s=%s: clf=%s', self.stock, self.r,
self.s, clf)
train_x = self.df[self.feature_columns_dict[feature_name]].values
test_x = self.df_test[self.feature_columns_dict[feature_name]].values
train_y = self.df['mid_price_indicator'].values
test_y = self.df_test['mid_price_indicator'].values
pca = self.get_pca(feature_name)
if pca:
train_x = pca.transform(train_x)
test_x = pca.transform(test_x)
if n_steps:
train_x, train_y = self.split_sequences(train_x,
train_y,
n_steps=n_steps)
test_x, test_y = self.split_sequences(test_x,
test_y,
n_steps=n_steps)
else:
train_x = np.reshape(train_x,
(train_x.shape[0], 1, train_x.shape[1]))
test_x = np.reshape(test_x, (test_x.shape[0], 1, test_x.shape[1]))
if should_validate:
scores_arrays, m = model.validate_model_lstm(
clf,
train_x,
train_y,
fit_kwargs=fit_kwargs,
compile_kwargs=compile_kwargs,
plot_name=plot_name,
class_weight=class_weight,
print_debug=False)
scores = self.get_mean_scores(scores_arrays)
else:
m = clf()
scores = model.train_model(m,
train_x,
train_y,
compile_kwargs=compile_kwargs,
fit_kwargs=fit_kwargs,
is_lstm=True,
class_weight=class_weight)
if not method:
method = 'lstm'
components_num = None
if pca:
components_num = pca.n_components_
res = {
**scores, 'stock': self.stock,
'kernel': method,
'features': feature_name,
'pca_components': components_num
}
# m = clf()
# model.train_model(
# m, train_x, train_y, compile_kwargs=compile_kwargs, fit_kwargs=fit_kwargs, is_lstm=True,
# class_weight=class_weight)
test_scores = model.test_model(m, test_x, test_y, is_lstm=True)
logger.info('Finished training %s %s', self.stock, {
**res,
**test_scores
})
if should_return_model:
return {**res, **test_scores, 'arch': m.to_json()}, m
else:
return {**res, **test_scores, 'arch': m.to_json()}
def get_score_for_clf(clf, df_test):
x_test = df_test[['queue_imbalance']]
y_test = df_test['mid_price_indicator'].values
return model.test_model(clf, x_test, y_test)
def get_score_for_clf(self, clf, df_test: pd.DataFrame,
feature_name: str) -> dict:
x_test = df_test[self.feature_columns_dict[feature_name]]
y_test = df_test['mid_price_indicator'].values
return model.test_model(clf, x_test, y_test)