Skip to content
Cannot retrieve contributors at this time
"""Extract LIEF features from PE files"""
from h2oaicore.transformer_utils import CustomTransformer
import datatable as dt
import numpy as np
class PEGeneralFeatures(CustomTransformer):
_modules_needed_by_name = ['lief==0.11.4']
_regression = True
_binary = True
_multiclass = True
_is_reproducible = True
_parallel_task = True # if enabled, params_base['n_jobs'] will be >= 1 (adaptive to system), otherwise 1
_can_use_gpu = True # if enabled, will use special job scheduler for GPUs
_can_use_multi_gpu = True # if enabled, can get access to multiple GPUs for single transformer (experimental)
_numeric_output = True
def get_default_properties():
return dict(col_type="text", min_cols=1, max_cols=1, relative_importance=1)
def do_acceptance_test():
return False
def fit_transform(self, X: dt.Frame, y: np.array = None):
return self.transform(X)
def load_pe(self, file_path):
with open(file_path, 'rb') as f:
bytez = bytearray(
return (bytez)
def get_general_features(self, file_path):
import lief
pe_bytez = self.load_pe(file_path)
lief_binary = lief.PE.parse(list(pe_bytez))
X = {'exports_count': len(lief_binary.exported_functions),
'imports_count': len(lief_binary.imported_functions),
'has_configuration': int(lief_binary.has_configuration),
'has_debug': int(lief_binary.has_debug),
'has_exceptions': int(lief_binary.has_exceptions),
'has_nx': int(lief_binary.has_nx),
'has_relocations': int(lief_binary.has_relocations),
'has_resources': int(lief_binary.has_resources),
'has_rich_header': int(lief_binary.has_rich_header),
'has_signature': int(lief_binary.has_signature),
'has_tls': int(lief_binary.has_tls),
'libraries_count': len(lief_binary.libraries),
'size': len(pe_bytez),
'symbols_count': len(lief_binary.symbols),
'virtual_size': lief_binary.virtual_size}
return X
X = {'exports_count': 0,
'imports_count': 0,
'has_configuration': 0,
'has_debug': 0,
'has_exceptions': 0,
'has_nx': 0,
'has_relocations': 0,
'has_resources': 0,
'has_rich_header': 0,
'has_signature': 0,
'has_tls': 0,
'libraries_count': 0,
'size': 0,
'symbols_count': 0,
'virtual_size': 0}
return X
def transform(self, X: dt.Frame):
import pandas as pd
ret_df = pd.DataFrame(
for x in X.to_pandas().values[:, 0]
self._output_feature_names = ['General_{}'.format(x) for x in ret_df.columns.to_list()]
self._feature_desc = self._output_feature_names
return ret_df