def test_boosting_type(boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
train_file = data_file('adult', 'train_small')
test_file = data_file('adult', 'test_small')
cd_file = data_file('adult', 'train.cd')
params = {
'--use-best-model': 'false',
'--loss-function': 'Logloss',
'-f': train_file,
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '10',
'-w': '0.03',
'-T': '4',
'-r': '0',
'-m': output_model_path,
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_queryrmse(boosting_type, qwise_loss):
output_model_path = yatest.common.test_output_path('model.bin')
test_error_path = yatest.common.test_output_path('test_error.tsv')
learn_error_path = yatest.common.test_output_path('learn_error.tsv')
predictions_path_learn = yatest.common.test_output_path('predictions_learn.tsv')
predictions_path_test = yatest.common.test_output_path('predictions_test.tsv')
learn_file = data_file('querywise', 'train')
cd_file = data_file('querywise', 'train.cd')
test_file = data_file('querywise', 'test')
params = {"--loss-function": qwise_loss,
"-f": learn_file,
"-t": test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '100',
'-T': '4',
'-r': '0',
'-m': output_model_path,
'--learn-err-log': learn_error_path,
'--test-err-log': test_error_path,
'--use-best-model': 'false'
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, learn_file, cd_file, predictions_path_learn)
apply_catboost(output_model_path, test_file, cd_file, predictions_path_test)
return [local_canonical_file(learn_error_path),
local_canonical_file(test_error_path),
local_canonical_file(predictions_path_learn),
local_canonical_file(predictions_path_test),
]
def test_overfit_detector_inc_to_dec(boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('adult', 'train.cd')
test_file = data_file('adult', 'test_small')
params = {
'--use-best-model': 'false',
'--loss-function': 'Logloss',
'-f': data_file('adult', 'train_small'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '2000',
'-T': '4',
'-r': '0',
'-m': output_model_path,
'-x': '1',
'-n': '8',
'-w': '0.5',
'--od-pval': '0.5',
'--od-type': 'IncToDec',
'--od-wait': '2',
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_all_targets(loss_function, boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
test_file = data_file('adult', 'test_small')
cd_file = data_file('adult', 'train.cd')
params = (
'--use-best-model',
'false',
'--loss-function',
loss_function,
'-f',
data_file('adult', 'train_small'),
'-t',
test_file,
'--column-description',
cd_file,
'--boosting-type',
boosting_type,
'-i',
'10',
'-w',
'0.03',
'-T',
'4',
'-r',
'0',
'-m',
output_model_path,
)
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def run_catboost(eval_path, model_path, train,
is_additional_query_weights):
fit_params = [
'--use-best-model',
'false',
'--loss-function',
'QueryRMSE',
'-f',
'quantized://' + data_file('querywise', train),
'-i',
'5',
'-T',
'4',
'-r',
'0',
'-m',
model_path,
'--eval-file',
eval_path,
]
if is_additional_query_weights:
fit_params += [
'--learn-group-weights',
data_file('querywise', 'train.group_weights'),
'--test-group-weights',
data_file('querywise', 'test.group_weights'),
]
fit_catboost_gpu(fit_params)
apply_catboost(model_path, data_file('querywise', 'test'),
data_file('querywise', 'train.cd.group_weight'),
eval_path)
def test_custom_priors(boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
test_file = data_file('adult', 'test_small')
cd_file = data_file('adult', 'train.cd')
params = (
'--use-best-model', 'false',
'--loss-function', 'Logloss',
'-f', data_file('adult', 'train_small'),
'-t', test_file,
'--column-description', cd_file,
'--boosting-type', boosting_type,
'-i', '10',
'-w', '0.03',
'-T', '4',
'-r', '0',
'-m', output_model_path,
'--ctr', 'Borders:Prior=-2:Prior=0:Prior=8/3:Prior=1:Prior=-1:Prior=3,'
'FeatureFreq:Prior=0',
'--per-feature-ctr', '4:Borders:Prior=0.444,FeatureFreq:Prior=0.444;'
'6:Borders:Prior=0.666,FeatureFreq:Prior=0.666;'
'8:Borders:Prior=-0.888:Prior=2/3,FeatureFreq:Prior=-0.888:Prior=0.888'
)
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def run_catboost(eval_path, model_path, cd_file,
is_additional_query_weights):
cd_file_path = data_file('querywise', cd_file)
fit_params = [
'--use-best-model',
'false',
'--loss-function',
'QueryRMSE',
'-f',
data_file('querywise', 'train'),
'--column-description',
cd_file_path,
'-i',
'5',
'-T',
'4',
'-r',
'0',
'-m',
model_path,
'--eval-file',
eval_path,
]
if is_additional_query_weights:
fit_params += [
'--learn-group-weights',
data_file('querywise', 'train.group_weights'),
'--test-group-weights',
data_file('querywise', 'test.group_weights'),
]
fit_catboost_gpu(fit_params)
apply_catboost(model_path, data_file('querywise', 'test'),
cd_file_path, eval_path)
def test_pairs_generation_with_max_pairs():
output_model_path = yatest.common.test_output_path('model.bin')
test_error_path = yatest.common.test_output_path('test_error.tsv')
learn_error_path = yatest.common.test_output_path('learn_error.tsv')
predictions_path_learn = yatest.common.test_output_path(
'predictions_learn.tsv')
predictions_path_test = yatest.common.test_output_path(
'predictions_test.tsv')
cd_file = data_file('querywise', 'train.cd')
learn_file = data_file('querywise', 'train')
test_file = data_file('querywise', 'test')
params = [
'--loss-function', 'PairLogit:max_pairs=30', '--eval-metric',
'PairAccuracy', '-f', learn_file, '-t', test_file,
'--column-description', cd_file, '--l2-leaf-reg', '0', '-i', '20',
'-T', '4', '-r', '0', '-m', output_model_path, '--learn-err-log',
learn_error_path, '--test-err-log', test_error_path,
'--use-best-model', 'false'
]
fit_catboost_gpu(params)
apply_catboost(output_model_path, learn_file, cd_file,
predictions_path_learn)
apply_catboost(output_model_path, test_file, cd_file,
predictions_path_test)
return [
local_canonical_file(learn_error_path, diff_tool=diff_tool()),
local_canonical_file(test_error_path, diff_tool=diff_tool()),
local_canonical_file(predictions_path_learn, diff_tool=diff_tool()),
local_canonical_file(predictions_path_test, diff_tool=diff_tool()),
]
def test_allow_writing_files_and_used_ram_limit(boosting_type, used_ram_limit):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('airlines_5K', 'cd')
params = (
'--use-best-model', 'false',
'--allow-writing-files', 'false',
'--used-ram-limit', used_ram_limit,
'--loss-function', 'Logloss',
'--max-ctr-complexity', '8',
'--depth', '10',
'-f', data_file('airlines_5K', 'train'),
'-t', data_file('airlines_5K', 'test'),
'--column-description', cd_file,
'--has-header',
'--boosting-type', boosting_type,
'-i', '20',
'-w', '0.03',
'-T', '4',
'-r', '0',
'-m', output_model_path,
'--eval-file', output_eval_path,
)
fit_catboost_gpu(params)
test_file = data_file('airlines_5K', 'test')
apply_catboost(output_model_path, test_file, cd_file,
output_eval_path, has_header=True)
return [local_canonical_file(output_eval_path)]
def test_bootstrap(boosting_type):
bootstrap_option = {
'no': {
'--bootstrap-type': 'No'
},
'bayes': {
'--bootstrap-type': 'Bayesian',
'--bagging-temperature': '0.0'
},
'bernoulli': {
'--bootstrap-type': 'Bernoulli',
'--subsample': '1.0'
}
}
test_file = data_file('adult', 'test_small')
cd_file = data_file('adult', 'train.cd')
params = {
'--use-best-model': 'false',
'--loss-function': 'Logloss',
'--random-strength': '0',
'-f': data_file('adult', 'train_small'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '10',
'-w': '0.03',
'-T': '4',
'-r': '0',
}
for bootstrap in bootstrap_option:
model_path = yatest.common.test_output_path('model_' + bootstrap +
'.bin')
eval_path = yatest.common.test_output_path('test_' + bootstrap +
'.eval')
model_option = {'-m': model_path}
run_params = combine_dicts(params, bootstrap_option[bootstrap],
model_option)
fit_catboost_gpu(run_params)
apply_catboost(model_path, test_file, cd_file, eval_path)
ref_eval_path = yatest.common.test_output_path('test_no.eval')
assert (filecmp.cmp(ref_eval_path,
yatest.common.test_output_path('test_bayes.eval')))
assert (filecmp.cmp(ref_eval_path,
yatest.common.test_output_path('test_bernoulli.eval')))
return [local_canonical_file(ref_eval_path)]
def test_ctr_type(ctr_type, boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('adult_crossentropy', 'train.cd')
test_file = data_file('adult_crossentropy', 'test_proba')
params = ('--use-best-model', 'false', '--loss-function', 'RMSE', '-f',
data_file('adult_crossentropy', 'train_proba'), '-t', test_file,
'--column-description', cd_file, '--boosting-type',
boosting_type, '-i', '3', '-T', '4', '-r', '0', '-m',
output_model_path, '--ctr', ctr_type)
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_train_on_binarized_equal_train_on_float(boosting_type, qwise_loss):
output_model_path = yatest.common.test_output_path('model.bin')
output_model_path_binarized = yatest.common.test_output_path('model_binarized.bin')
test_error_path = yatest.common.test_output_path('test_error.tsv')
learn_error_path = yatest.common.test_output_path('learn_error.tsv')
borders_file = yatest.common.test_output_path('borders.tsv')
borders_file_output = borders_file + '.out'
predictions_path_learn = yatest.common.test_output_path('predictions_learn.tsv')
predictions_path_learn_binarized = yatest.common.test_output_path('predictions_learn_binarized.tsv')
predictions_path_test = yatest.common.test_output_path('predictions_test.tsv')
predictions_path_test_binarized = yatest.common.test_output_path('predictions_test_binarized.tsv')
learn_file = data_file('querywise', 'train')
cd_file = data_file('querywise', 'train.cd')
test_file = data_file('querywise', 'test')
params = {"--loss-function": qwise_loss,
"-f": learn_file,
"-t": test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '100',
'-T': '4',
'-r': '0',
'-m': output_model_path,
'--learn-err-log': learn_error_path,
'--test-err-log': test_error_path,
'--use-best-model': 'false',
'--output-borders-file': borders_file_output,
}
params_binarized = dict(params)
params_binarized['--input-borders-file'] = borders_file_output
params_binarized['--output-borders-file'] = borders_file
params_binarized['-m'] = output_model_path_binarized
fit_catboost_gpu(params)
apply_catboost(output_model_path, learn_file, cd_file, predictions_path_learn)
apply_catboost(output_model_path, test_file, cd_file, predictions_path_test)
# learn_error_path and test_error_path already exist after first fit_catboost_gpu() call
# and would be automatically marked as input_data for YT operation,
# which will lead to error, because input files are available only for reading.
# That's why we explicitly drop files from input_data and implicitly add them to output_data.
fit_catboost_gpu(params_binarized, input_data={learn_error_path: None, test_error_path: None})
apply_catboost(output_model_path_binarized, learn_file, cd_file, predictions_path_learn_binarized)
apply_catboost(output_model_path_binarized, test_file, cd_file, predictions_path_test_binarized)
assert (filecmp.cmp(predictions_path_learn, predictions_path_learn_binarized))
assert (filecmp.cmp(predictions_path_test, predictions_path_test_binarized))
return [local_canonical_file(learn_error_path),
local_canonical_file(test_error_path),
local_canonical_file(predictions_path_test),
local_canonical_file(predictions_path_learn),
local_canonical_file(borders_file)]
def test_weighted_pool_leaf_estimation_method(boosting_type, leaf_estimation):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('adult_weight', 'train.cd')
test_file = data_file('adult_weight', 'test_weight')
params = {
'--use-best-model': 'false',
'--loss-function': 'Logloss',
'-f': data_file('adult_weight', 'train_weight'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '10',
'-T': '4',
'-r': '0',
'--leaf-estimation-method': leaf_estimation,
'-m': output_model_path,
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_crossentropy(boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('adult_crossentropy', 'train.cd')
test_file = data_file('adult_crossentropy', 'test_proba')
params = {
'--loss-function': 'CrossEntropy',
'-f': data_file('adult_crossentropy', 'train_proba'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '10',
'-w': '0.03',
'-T': '4',
'-r': '0',
'-m': output_model_path,
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_nan_mode(nan_mode, boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
test_file = data_file('adult_nan', 'test_small')
cd_file = data_file('adult_nan', 'train.cd')
params = {
'--use-best-model': 'false',
'-f': data_file('adult_nan', 'train_small'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': boosting_type,
'-i': '20',
'-T': '4',
'-r': '0',
'-m': output_model_path,
'--nan-mode': nan_mode
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_fold_len_mult():
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
cd_file = data_file('adult_not_binarized', 'train.cd')
test_file = data_file('adult_not_binarized', 'test_small')
params = {
'--use-best-model': 'false',
'--loss-function': 'Logloss',
'-f': data_file('adult_not_binarized', 'train_small'),
'-t': test_file,
'--column-description': cd_file,
'--boosting-type': 'Ordered',
'-i': '10',
'-w': '0.03',
'-T': '4',
'-r': '0',
'--fold-len-multiplier': 1.2,
'-m': output_model_path,
}
fit_catboost_gpu(params)
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
def test_quantized_pool(loss_function, boosting_type):
output_model_path = yatest.common.test_output_path('model.bin')
output_eval_path = yatest.common.test_output_path('test.eval')
quantized_train_file = 'quantized://' + data_file('quantized_adult',
'train.qbin')
quantized_test_file = 'quantized://' + data_file('quantized_adult',
'test.qbin')
params = (
'--use-best-model',
'false',
'--loss-function',
loss_function,
'-f',
quantized_train_file,
'-t',
quantized_test_file,
'--boosting-type',
boosting_type,
'-i',
'10',
'-w',
'0.03',
'-T',
'4',
'-r',
'0',
'-m',
output_model_path,
)
fit_catboost_gpu(params)
cd_file = data_file('quantized_adult', 'pool.cd')
test_file = data_file('quantized_adult', 'test_small.tsv')
apply_catboost(output_model_path, test_file, cd_file, output_eval_path)
return [local_canonical_file(output_eval_path)]
