Change parameters type for custom tuning

[ppyarpe]

I am currently using FLAML with custom tunning for a DPBoost (https://github.com/QinbinLi/DPBoost) implementation.

DPBoost overrides lightgbm and hence to use it ones uses the same functions as LightGBM.

When I run lgb.train(params,data) I get the following error:

File "", line 1, in
model=lgb.train(params,data)

File "C:\Users\Rita\AppData\Roaming\Python\Python37\site-packages\lightgbm\engine.py", line 156, in train
if num_boost_round <= 0:

TypeError: '<=' not supported between instances of 'Integer' and 'int'

Params:
{'num_leaves': <flaml.tune.sample.Integer at 0x185a255ef28>,
'max_depth': <flaml.tune.sample.Integer at 0x185a255eda0>,
'learning_rate': <flaml.tune.sample.Float at 0x185a255e630>,
'num_iterations': <flaml.tune.sample.Integer at 0x185a255e780>,
'my_n_trees': <flaml.tune.sample.Integer at 0x185a255edd8>,
'lambda_l2': <flaml.tune.sample.Float at 0x185a253a048>,
'bagging_freq': <flaml.tune.sample.Integer at 0x185a3c11c50>,
'bagging_fraction': <flaml.tune.sample.Float at 0x185a253a160>,
'max_bin': <flaml.tune.sample.Integer at 0x185a253a0b8>,
'inner_boost_round': <flaml.tune.sample.Integer at 0x185a253a1d0>,
'total_budget': 1,
'verbose': -2,
'boosting_type': 'gbdt',
'objective': 'regression',
'metric': 'binary_error',
'boost_method': 'DPBoost_2level',
'high_level_boost_round': 1,
'balance_partition': 0,
'geo_clip': 1,
'boost_from_average': False}

where the flaml.tune objects come from the configurations and search space.

And data is a lightgbm Dataset.

Is there any way to change the flaml.tune objects into different types in order to feed into the lgb.train function from DPBoost ?

When I try int(params['num_leaves']) ,for example, I get the following error:

Traceback (most recent call last):

File "", line 1, in
int(params['num_leaves'])

TypeError: int() argument must be a string, a bytes-like object or a number, not 'Integer'

[sonichi]

The params in your code seems to be a search space instead of a specific config. Could you share the code that causes the error?

[qingyun-wu]

Hi @ppyarpe, with the provided code above, I do not get any error. Just some warnings from LightGBM such as[LightGBM] [Warning] Unknown parameter: my_n_trees. I am testing with flamlv0.10.0 and python 3.8. Do you still have the error you described?

Thank you!

[ppyarpe]

Hi @qingyun-wu, yes I am still getting the same error when using DPBoost. However installing flaml after DPBoost (which needs to be done using Github specifications) will rewrite DPBoost and hence revert to LightGBM in which case I don't get the error.

To double check if DPBoost is working one should have different metric results for different total_budget

[qingyun-wu]

Hi @ppyarpe , I am trying install DPBoost following the instruction from the link you provided, but I get the following error when installing it.

  File "setup.py", line 337, in <module>
    'Topic :: Scientific/Engineering :: Artificial Intelligence'])
  File "/usr/lib/python3.7/site-packages/setuptools/__init__.py", line 165, in setup
    return distutils.core.setup(**attrs)
  File "/usr/lib64/python3.7/distutils/core.py", line 148, in setup
    dist.run_commands()
  File "/usr/lib64/python3.7/distutils/dist.py", line 966, in run_commands
    self.run_command(cmd)
  File "/usr/lib64/python3.7/distutils/dist.py", line 985, in run_command
    cmd_obj.run()
  File "setup.py", line 261, in run
    openmp_include_dir=self.openmp_include_dir, openmp_library=self.openmp_library)
  File "setup.py", line 200, in compile_cpp
    silent_call(cmake_cmd, raise_error=True, error_msg='Please install CMake and all required dependencies first')
  File "setup.py", line 86, in silent_call
    raise Exception("\n".join((error_msg, LOG_NOTICE)))
Exception: Please install CMake and all required dependencies first
The full version of error log was saved into /root/LightGBM_compilation.log

I tried re-installing cmake and gcc, but still getting the same error. Any idea how should I resolve this error during DPBoost installation?

And about To double check if DPBoost is working one should have different metric results for different total_budget: I am not sure how should I interpret it. Do you mean after successfully installing DPBoost, I should revise the code such that there are different metric results for different total_budget to test if DPBoost is working? Can you revise your code to reflect it such that I can directly run your code?

Thank you!

[qingyun-wu]

@ppyarpe, like @sonichi said, the params in your code seem to be a flaml search space instead of a specific config. Since I cannot test your code directly now, you might want to double-check whether the input params in model=lgb.train(params, data), and the config in LGBMEstimator(**config).params are properly obtained. It will also be helpful if you can share your whole error stack trace.
Thank you!

[ppyarpe]

@qingyun-wu, regarding the installation of DPBoost I am not sure when I install it in Windows I have downloaded Visual Studio 16 and that has worked and when installing it on a mac I have installed cmake, gcc version 7 and libomp.

Thank you I will double check the params input ! :)