def prepDataSet(csv_filename, feature_set=None, dataset_name='generic dataset',
ddg_cutoff=0.0, truncate=False):
'''
prepares a data set object from a CSV file, under the conventions of this project:
- the CSV is indexed by PDBID and residue number (columns 0,1)
- the last column contains label-related data, mostly ddG values of residues.
- all other columns are feature columns.
The function reads the columns into a TreeDict structure, such that each component
(normalized feature data, labels, PDB identifiers, columns used) is accessible as
an attribute.
``dataset_name`` is optional, giving the TreeDict a name.
Optional argument ``features`` directs the function which features to select from
the table. By default, all features are selected.
'''
dataset = TreeDict(dataset_name)
dataset.csv_filename = os.path.abspath(csv_filename)
dataset.is_bound = (csv_filename.find('unbound') == -1)
dataset._df = cached_csv_df(csv_filename, index_col=[0,1],
true_values=['True'],
false_values=['False'],
)
if truncate:
dataset._df = dataset._df[:DEBUG_DATASET_SIZE]
if feature_set is None:
cols = dataset._df.columns[:-1]
dataset.feature_set = FeatureSet(cols, cols)
else:
dataset.feature_set = feature_set
all_feature_data_df = dataset._df.ix[:,dataset.feature_set.all_features]
dataset.feature_data_df = all_feature_data_df.ix[:,dataset.feature_set.features]
#dataset.X = dataset.feature_data_df.values
dataset.X = sklearn.preprocessing.scale(
dataset.feature_data_df.values.astype(float))
dataset.label_data_df = dataset._df.ix[:,-1]
dataset.ddg_cutoff = ddg_cutoff
dataset.y = dataset.label_data_df.values > dataset.ddg_cutoff
# sanity checks
assert dataset.X.shape[0] == len(dataset.y)
dataset.pdbs = dataset.feature_data_df.index.get_level_values(0)
return dataset
