def cv_on_dataset(dataset,
feature_param,
model_param,
ax,
result_store,
contentid_groups,
logger=None,
aggregate_method=np.mean):
assets = read_dataset(dataset)
kfold = construct_kfold_list(assets, contentid_groups)
fassembler = FeatureAssembler(
feature_dict=feature_param.feature_dict,
feature_option_dict=None,
assets=assets,
logger=logger,
delete_workdir=True,
result_store=result_store,
optional_dict=None,
optional_dict2=None,
parallelize=True,
fifo_mode=True,
# parallelize=False, fifo_mode=False, # VQM
)
fassembler.run()
results = fassembler.results
for result in results:
result.set_score_aggregate_method(aggregate_method)
model_class = TrainTestModel.find_subclass(model_param.model_type)
# run nested kfold cv for each combintation
cv_output = ModelCrossValidation.run_kfold_cross_validation(
model_class,
model_param.model_param_dict,
results,
kfold,
logger=logger,
)
print('Feature parameters: {}'.format(feature_param.feature_dict))
print('Model type: {}'.format(model_param.model_type))
print('Model parameters: {}'.format(model_param.model_param_dict))
print('Stats: {}'.format(
model_class.format_stats_for_print(cv_output['aggr_stats'])))
if ax is not None:
model_class.plot_scatter(ax,
cv_output['aggr_stats'],
content_ids=cv_output['contentids'])
ax.set_xlabel('True Score')
ax.set_ylabel("Predicted Score")
ax.grid()
ax.set_title("Dataset: {dataset}, Model: {model},\n{stats}".format(
dataset=dataset.dataset_name,
model=model_param.model_type,
stats=model_class.format_stats_for_plot(cv_output['aggr_stats'])))
return assets, cv_output
def test_run_kfold_cross_validation_with_list_input(self):
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {
'norm_type': 'normalize',
'n_estimators': 10,
'random_state': 0
}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features,
[[0, 3, 8], [2, 1, 5], [4, 6, 7]])
self.assertAlmostEqual(output['aggr_stats']['SRCC'],
0.18333333333333335,
places=4)
self.assertAlmostEqual(output['aggr_stats']['PCC'],
0.35513638509959689,
places=4)
self.assertAlmostEqual(output['aggr_stats']['KENDALL'],
0.1111111111111111,
places=3)
self.assertAlmostEqual(output['aggr_stats']['RMSE'],
1.2740400878438387,
places=3)
def test_run_kfold_cross_validation_libsvmnusvr(self):
train_test_model_class = LibsvmNusvrTrainTestModel
model_param = {'norm_type': 'normalize'}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEqual(output['aggr_stats']['SRCC'], 0.31666666666666665, places=4)
self.assertAlmostEqual(output['aggr_stats']['PCC'], 0.33103132578536021, places=4)
self.assertAlmostEqual(output['aggr_stats']['KENDALL'], 0.27777777777777779, places=4)
self.assertAlmostEqual(output['aggr_stats']['RMSE'], 1.2855099934718619, places=4)
def test_run_kfold_cross_validation_extratrees(self):
train_test_model_class = SklearnExtraTreesTrainTestModel
model_param = {'norm_type': 'normalize', 'random_state': 0, 'n_estimators': 10}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEqual(output['aggr_stats']['SRCC'], 0.17320508075688773, places=4)
self.assertAlmostEqual(output['aggr_stats']['PCC'], 0.33023719320146966, places=4)
self.assertAlmostEqual(output['aggr_stats']['KENDALL'], 0.14907119849998599, places=4)
self.assertAlmostEqual(output['aggr_stats']['RMSE'], 1.3279056191361394, places=4)
def cv_on_dataset(dataset, feature_param, model_param, ax, result_store,
contentid_groups, logger=None, aggregate_method=np.mean):
assets = read_dataset(dataset)
kfold = construct_kfold_list(assets, contentid_groups)
fassembler = FeatureAssembler(
feature_dict=feature_param.feature_dict,
feature_option_dict=None,
assets=assets,
logger=logger,
delete_workdir=True,
result_store=result_store,
optional_dict=None,
optional_dict2=None,
parallelize=True, fifo_mode=True,
# parallelize=False, fifo_mode=False, # VQM
)
fassembler.run()
results = fassembler.results
for result in results:
result.set_score_aggregate_method(aggregate_method)
model_class = TrainTestModel.find_subclass(model_param.model_type)
# run nested kfold cv for each combintation
cv_output = ModelCrossValidation.run_kfold_cross_validation(
model_class,
model_param.model_param_dict,
results,
kfold,
logger=logger,
)
print 'Feature parameters: {}'.format(feature_param.feature_dict)
print 'Model type: {}'.format(model_param.model_type)
print 'Model parameters: {}'.format(model_param.model_param_dict)
print 'Stats: {}'.format(model_class.format_stats(cv_output['aggr_stats']))
if ax is not None:
model_class.plot_scatter(ax, cv_output['aggr_stats'], cv_output['contentids'])
ax.set_xlabel('True Score')
ax.set_ylabel("Predicted Score")
ax.grid()
ax.set_title("Dataset: {dataset}, Model: {model},\n{stats}".format(
dataset=dataset.dataset_name,
model=model_param.model_type,
stats=model_class.format_stats(cv_output['aggr_stats'])
))
return assets, cv_output
def test_run_kfold_cross_validation_libsvmnusvr(self):
print "test k-fold cross validation on libsvmnusvr..."
train_test_model_class = LibsvmNusvrTrainTestModel
model_param = {'norm_type': 'normalize'}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.31666666666666665, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.33103132578536021, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.27777777777777779, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.2855099934718619, places=4)
def test_run_kfold_cross_validation_extratrees(self):
print "test k-fold cross validation on extra trees..."
train_test_model_class = SklearnExtraTreesTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.17320508075688773, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.33023719320146966, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.14907119849998599, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.3279056191361394, places=4)
def test_run_kfold_cross_validation_randomforest(self):
print "test k-fold cross validation on random forest..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.28452131897694583, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.1689046198483892, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.084515425472851652, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.344683833136588, places=4)
def test_run_kfold_cross_validation_randomforest(self):
print "test k-fold cross validation on random forest..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.28452131897694583, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.1689046198483892, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.084515425472851652, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.344683833136588, places=4)
def test_run_kfold_cross_validation_with_list_input(self):
print "test k-fold cross validation with list input..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, self.features,
[[0, 3, 8], [2, 1, 5], [4, 6, 7]])
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.18333333333333335, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.35513638509959689, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.1111111111111111, places=3)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.2740400878438387, places=3)
