def test_sample_model_param_list(self):
import random
random.seed(0)
model_param_search_range = {'norm_type':['normalize', 'clip_0to1'],
'n_estimators':[10, 50], 'random_state': [0]}
dicts = ModelCrossValidation._sample_model_param_list(
model_param_search_range, 4)
expected_dicts = [
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':10, 'random_state':0},
{'norm_type':'normalize', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
]
self.assertEquals(dicts, expected_dicts)
model_param_search_range = {'norm_type':['normalize', 'clip_0to1'],
'n_estimators':{'low':10, 'high':50, 'decimal':0},
'random_state': [0]}
dicts = ModelCrossValidation._sample_model_param_list(
model_param_search_range, 4)
expected_dicts = [
{'norm_type':'clip_0to1', 'n_estimators':21, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':20, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':42, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':39, 'random_state':0},
]
self.assertEquals(dicts, expected_dicts)
def test_sample_model_param_list(self):
import random
random.seed(0)
model_param_search_range = {'norm_type':['normalize', 'clip_0to1'],
'n_estimators':[10, 50], 'random_state': [0]}
dicts = ModelCrossValidation._sample_model_param_list(
model_param_search_range, 4)
expected_dicts = [
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':10, 'random_state':0},
{'norm_type':'normalize', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
]
self.assertEquals(dicts, expected_dicts)
model_param_search_range = {'norm_type':['normalize', 'clip_0to1'],
'n_estimators':{'low':10, 'high':50, 'decimal':0},
'random_state': [0]}
dicts = ModelCrossValidation._sample_model_param_list(
model_param_search_range, 4)
expected_dicts = [
{'norm_type':'clip_0to1', 'n_estimators':21, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':20, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':42, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':39, 'random_state':0},
]
self.assertEquals(dicts, expected_dicts)
def test_run_nested_kfold_cross_validation_with_list_input(self):
print "test nested k-fold cross validation with list input..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['none'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features,
[[0, 3, 2], [8, 6, 5], [4, 1, 7]])
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.26666666666666666, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.15272340058922063, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.22222222222222221, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.452887116343635, places=4)
expected_top_model_param = {'norm_type':'none',
'n_estimators':10,
'max_depth':None,
'random_state':0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_cross_validation(self):
print "test cross validation..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type': 'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
indices_train = range(250)
indices_test = range(250, 300)
output = ModelCrossValidation.run_cross_validation(
train_test_model_class, model_param, feature_df, indices_train,
indices_test)
self.assertAlmostEquals(output['stats']['SRCC'],
0.93493301443051136,
places=4)
self.assertAlmostEquals(output['stats']['PCC'],
0.9413390374529329,
places=4)
self.assertAlmostEquals(output['stats']['KENDALL'],
0.78029280419726044,
places=4)
self.assertAlmostEquals(output['stats']['RMSE'],
0.32357946626958406,
places=4)
self.assertEquals(output['model'].TYPE, "RANDOMFOREST")
def test_run_nested_kfold_cross_validation_randomforest(self):
print "test nested k-fold cross validation on random forest..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.40167715620274708, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.11009919053282299, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.14085904245475275, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.3681348274719265, places=4)
expected_top_model_param = {'norm_type':'normalize',
'n_estimators':10,
'max_depth':None,
'random_state':0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_kfold_cross_validation_with_list_input(self):
print "test k-fold cross validation with list input..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type': 'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
feature_df = feature_df[:200]
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df,
[range(0, 50), range(130, 200),
range(50, 130)])
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.90636761259756715)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.90819953685397914)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.72937284548325965)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.49899297305829415)
def test_run_nested_kfold_cross_validation_libsvmnusvr(self):
print "test nested k-fold cross validation on libsvmnusvr..."
train_test_model_class = LibsvmNusvrTrainTestModel
model_param_search_range = \
{'norm_type':['normalize', 'clip_0to1', 'clip_minus1to1'],
'kernel':['rbf'],
'nu': [0.5],
'C': [1, 2],
'gamma': [0.0]
}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.30962614123961751, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], -0.1535643705229309, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.14085904245475275, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.5853397658781734, places=4)
expected_top_model_param = {'norm_type':'clip_0to1',
'kernel':'rbf',
'nu':0.5,
'C':1,
'gamma':0.0,
}
expected_top_ratio = 1.0
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_nested_kfold_cross_validation_randomforest(self):
print "test nested k-fold cross validation on random forest..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.40167715620274708, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.11009919053282299, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.14085904245475275, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.3681348274719265, places=4)
expected_top_model_param = {'norm_type':'normalize',
'n_estimators':10,
'max_depth':None,
'random_state':0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_nested_kfold_cross_validation_libsvmnusvr(self):
print "test nested k-fold cross validation on libsvmnusvr..."
train_test_model_class = LibsvmnusvrTrainTestModel
model_param_search_range = \
{'norm_type':['normalize', 'clip_0to1', 'clip_minus1to1'],
'kernel':['rbf'],
'nu': [0.5, 1.0],
'C': [1, 2],
'gamma': [0.0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.93704238362264514)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.94734024567912978)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.77785381654919195)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.34039563991411448)
expected_top_model_param = {'norm_type':'clip_0to1',
'kernel':'rbf',
'nu':1.0,
'C':1,
'gamma':0.0,
}
expected_top_ratio = 0.5
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_kfold_cross_validation_randomforest(self):
print "test k-fold cross validation on random forest..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type': 'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.92695443548602008,
places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.93189074441713937,
places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.76031309571294092,
places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.40381451586590256,
places=4)
def test_run_nested_kfold_cross_validation_with_list_input(self):
print "test nested k-fold cross validation with list input..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['none'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features,
[[0, 3, 2], [8, 6, 5], [4, 1, 7]])
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.26666666666666666, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.15272340058922063, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.22222222222222221, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.452887116343635, places=4)
expected_top_model_param = {'norm_type':'none',
'n_estimators':10,
'max_depth':None,
'random_state':0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_nested_kfold_cross_validation_randomforest(self):
print "test nested k-fold cross validation on random forest..."
train_test_model_class = RandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.92805802153293737)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.94066838465382363)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.76196220071567478)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.37660623901376861)
expected_top_model_param = {'norm_type':'normalize',
'n_estimators':90,
'max_depth':None,
'random_state':0
}
expected_top_ratio = 0.5
# self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_unroll_dict_of_lists(self):
model_param_search_range = {
'norm_type': ['normalize', 'clip_0to1'],
'n_estimators': [10, 50],
'random_state': [0]
}
dicts = ModelCrossValidation._unroll_dict_of_lists(
model_param_search_range)
expected_dicts = [
{
'norm_type': 'normalize',
'n_estimators': 10,
'random_state': 0
},
{
'norm_type': 'clip_0to1',
'n_estimators': 10,
'random_state': 0
},
{
'norm_type': 'normalize',
'n_estimators': 50,
'random_state': 0
},
{
'norm_type': 'clip_0to1',
'n_estimators': 50,
'random_state': 0
},
]
self.assertEquals(dicts, expected_dicts)
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'}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.92387451180595015,
places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.92481147926825724,
places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.75416215405673581,
places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.42231775639097513,
places=4)
def test_run_cross_validation(self):
print "test cross validation..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {'norm_type': 'normalize', 'random_state': 0}
indices_train = range(9)
indices_test = range(9)
output = ModelCrossValidation.run_cross_validation(
train_test_model_class, model_param, self.features, indices_train,
indices_test)
self.assertAlmostEquals(output['stats']['SRCC'],
0.93333333333333324,
places=4)
self.assertAlmostEquals(output['stats']['PCC'],
0.97754442316039469,
places=4)
self.assertAlmostEquals(output['stats']['KENDALL'],
0.83333333333333337,
places=4)
self.assertAlmostEquals(output['stats']['RMSE'],
0.17634739353518517,
places=4)
self.assertEquals(output['model'].TYPE, "RANDOMFOREST")
def test_run_nested_kfold_cross_validation_libsvmnusvr(self):
print "test nested k-fold cross validation on libsvmnusvr..."
train_test_model_class = LibsvmNusvrTrainTestModel
model_param_search_range = \
{'norm_type':['normalize', 'clip_0to1', 'clip_minus1to1'],
'kernel':['rbf'],
'nu': [0.5],
'C': [1, 2],
'gamma': [0.0]
}
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, self.features, 3)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.30962614123961751, places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'], -0.1535643705229309, places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.14085904245475275, places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 1.5853397658781734, places=4)
expected_top_model_param = {'norm_type':'clip_0to1',
'kernel':'rbf',
'nu':0.5,
'C':1,
'gamma':0.0,
}
expected_top_ratio = 1.0
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_nested_kfold_cross_validation_with_list_input(self):
print "test nested k-fold cross validation with list input..."
train_test_model_class = RandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
feature_df = feature_df[:200]
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df,
[range(0,50), range(130, 200), range(50, 130)]
)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.92549459243170684)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.93070443071372855)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.76385104263763215)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.43223946862572299)
expected_top_model_param = {'norm_type':'normalize',
'n_estimators':90,
'max_depth':3,
'random_state':0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
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 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_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_unroll_dict_of_lists(self):
model_param_search_range = {'norm_type':['normalize', 'clip_0to1'],
'n_estimators':[10, 50], 'random_state': [0]}
dicts = ModelCrossValidation._unroll_dict_of_lists(model_param_search_range)
expected_dicts = [
{'norm_type':'normalize', 'n_estimators':10, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':10, 'random_state':0},
{'norm_type':'normalize', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
]
self.assertEquals(dicts, expected_dicts)
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_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_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_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)
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)
def test_find_most_frequent_dict(self):
dicts = [
{'norm_type':'normalize', 'n_estimators':10, 'random_state':0},
{'norm_type':'normalize', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':10, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
]
dict, count = ModelCrossValidation._find_most_frequent_dict(dicts)
expected_dict = {'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0}
expected_count = 2
self.assertEquals(dict, expected_dict)
self.assertEquals(count, expected_count)
def test_find_most_frequent_dict(self):
dicts = [
{'norm_type':'normalize', 'n_estimators':10, 'random_state':0},
{'norm_type':'normalize', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':10, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
{'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0},
]
dict, count = ModelCrossValidation._find_most_frequent_dict(dicts)
expected_dict = {'norm_type':'clip_0to1', 'n_estimators':50, 'random_state':0}
expected_count = 2
self.assertEquals(dict, expected_dict)
self.assertEquals(count, expected_count)
def test_run_nested_kfold_cross_validation_libsvmnusvr(self):
print "test nested k-fold cross validation on libsvmnusvr..."
train_test_model_class = LibsvmnusvrTrainTestModel
model_param_search_range = \
{'norm_type':['normalize', 'clip_0to1', 'clip_minus1to1'],
'kernel':['rbf'],
'nu': [0.5, 1.0],
'C': [1, 2],
'gamma': [0.0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.93704238362264514,
places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.94734024567912978,
places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.77785381654919195,
places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.34039563991411448,
places=4)
expected_top_model_param = {
'norm_type': 'clip_0to1',
'kernel': 'rbf',
'nu': 1.0,
'C': 1,
'gamma': 0.0,
}
expected_top_ratio = 0.5
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_nested_kfold_cross_validation_with_list_input(self):
print "test nested k-fold cross validation with list input..."
train_test_model_class = RandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
feature_df = feature_df[:200]
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df,
[range(0, 50), range(130, 200),
range(50, 130)])
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.92549459243170684,
places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.93070443071372855,
places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.76385104263763215,
places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.43223946862572299,
places=4)
expected_top_model_param = {
'norm_type': 'normalize',
'n_estimators': 90,
'max_depth': 3,
'random_state': 0
}
expected_top_ratio = 0.6666666666666666
self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_kfold_cross_validation_randomforest(self):
print "test k-fold cross validation on random forest..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.92695443548602008)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.93189074441713937)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.76031309571294092)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.40381451586590256)
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'}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.92387451180595015)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.92481147926825724)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.75416215405673581)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.42231775639097513)
def test_run_cross_validation(self):
print "test cross validation..."
train_test_model_class = SklearnRandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
indices_train = range(9)
indices_test = range(9)
output = ModelCrossValidation.run_cross_validation(
train_test_model_class, model_param, self.features,
indices_train, indices_test)
self.assertAlmostEquals(output['stats']['SRCC'], 0.93333333333333324, places=4)
self.assertAlmostEquals(output['stats']['PCC'], 0.97754442316039469, places=4)
self.assertAlmostEquals(output['stats']['KENDALL'], 0.83333333333333337, places=4)
self.assertAlmostEquals(output['stats']['RMSE'], 0.17634739353518517, places=4)
self.assertEquals(output['model'].TYPE, "RANDOMFOREST")
def test_run_nested_kfold_cross_validation_randomforest(self):
print "test nested k-fold cross validation on random forest..."
train_test_model_class = RandomForestTrainTestModel
model_param_search_range = \
{'norm_type':['normalize'],
'n_estimators':[10, 90],
'max_depth':[None, 3],
'random_state': [0]
}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(
eval(open(feature_df_file, "r").read()))
output = ModelCrossValidation.run_nested_kfold_cross_validation(
train_test_model_class, model_param_search_range, feature_df, 6)
self.assertAlmostEquals(output['aggr_stats']['SRCC'],
0.92805802153293737,
places=4)
self.assertAlmostEquals(output['aggr_stats']['PCC'],
0.94066838465382363,
places=4)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'],
0.76196220071567478,
places=4)
self.assertAlmostEquals(output['aggr_stats']['RMSE'],
0.37660623901376861,
places=4)
expected_top_model_param = {
'norm_type': 'normalize',
'n_estimators': 90,
'max_depth': None,
'random_state': 0
}
expected_top_ratio = 0.5
# self.assertEquals(output['top_model_param'], expected_top_model_param)
self.assertEquals(output['top_ratio'], expected_top_ratio)
def test_run_kfold_cross_validation_with_list_input(self):
print "test k-fold cross validation with list input..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
feature_df = feature_df[:200]
output = ModelCrossValidation.run_kfold_cross_validation(
train_test_model_class, model_param, feature_df,
[range(0,50), range(130, 200), range(50, 130)])
self.assertAlmostEquals(output['aggr_stats']['SRCC'], 0.90636761259756715)
self.assertAlmostEquals(output['aggr_stats']['PCC'], 0.90819953685397914)
self.assertAlmostEquals(output['aggr_stats']['KENDALL'], 0.72937284548325965)
self.assertAlmostEquals(output['aggr_stats']['RMSE'], 0.49899297305829415)
def test_run_cross_validation(self):
print "test cross validation..."
train_test_model_class = RandomForestTrainTestModel
model_param = {'norm_type':'normalize', 'random_state': 0}
feature_df_file = config.ROOT + \
"/python/test/resource/sample_feature_extraction_results.json"
feature_df = pd.DataFrame.from_dict(eval(open(feature_df_file, "r").read()))
indices_train = range(250)
indices_test = range(250, 300)
output = ModelCrossValidation.run_cross_validation(
train_test_model_class, model_param, feature_df,
indices_train, indices_test)
self.assertAlmostEquals(output['stats']['SRCC'], 0.93493301443051136)
self.assertAlmostEquals(output['stats']['PCC'], 0.9413390374529329)
self.assertAlmostEquals(output['stats']['KENDALL'], 0.78029280419726044)
self.assertAlmostEquals(output['stats']['RMSE'], 0.32357946626958406)
self.assertEquals(output['model'].TYPE, "RANDOMFOREST")