This code implements PBCTs based on its original proposal by Pliakos, Geurts and Vens in 20181. Functionality has been extended to n-dimensional interaction tensors, where n instances of n different classes would interact or not for each database instance.
1Pliakos, Konstantinos, Pierre Geurts, and Celine Vens. "Global multi-output decision trees for interaction prediction." Machine Learning 107.8 (2018): 1257-1281.
The package is available at PyPI and can be installed by the usual pip command:
$ pip install pbct
Local installation can be done either by providing the --user flag to the above command or by cloning this repo and issuing pip afterwards, for example:
$ git clone https://github.com/pedroilidio/PCT
$ cd PCT
$ pip install -e .
Where the -e option installs it as symbolic links to the local cloned repository, so that changes in it will reflect on the installation directly.
Usage and input/output examples are provided in the examples folder.
We provide a command-line utility to use PBCT models, that shows the following information when the --help option is used.
$ PBCT --help
usage: PBCT [-h] [--config CONFIG] [--XX XX [XX ...]]
[--XX_names XX_NAMES [XX_NAMES ...]]
[--XX_col_names XX_COL_NAMES [XX_COL_NAMES ...]] [--Y Y]
[--path_model PATH_MODEL] [--max_depth MAX_DEPTH]
[--min_samples_leaf MIN_SAMPLES_LEAF] [--verbose]
[--outdir OUTDIR] [--k K [K ...]] [--diag]
[--test_size TEST_SIZE [TEST_SIZE ...]]
[--train_size TRAIN_SIZE [TRAIN_SIZE ...]] [--njobs NJOBS]
[--random_state RANDOM_STATE]
[{fit,predict,train_test,xval}]
Fit a PBCT model to data or use a trained model to predict new results. Input
files and options may be provided either with command-line options or by a
json config file (see --config option).
positional arguments:
{fit,predict,train_test,xval}
fit: Use input data to train a PBCT. predict: Predict
interaction between input instances. train_test: Split
data between the 4 train/test sets, train and test a
PBCT. xval: run a 2D k-fold cross validation with the
given data. (default: None)
optional arguments:
-h, --help show this help message and exit
--config CONFIG Load options from json file. File example: {
"path_model": "/path/to/save/model.json", "fit":
"true", "XX": ["/path/to/X1.csv", "/path/to/X2.csv"],
"Y": "/path/to/Y.csv", }. Multiple dicts in a list
will cause this script to run multiple times.
(default: None)
--XX XX [XX ...] Paths to .csv files containing rows of numerical
attributes for each axis' instance. (default: None)
--XX_names XX_NAMES [XX_NAMES ...]
Paths to files containing string identifiers for each
instance for each axis, being one file for each axis.
(default: None)
--XX_col_names XX_COL_NAMES [XX_COL_NAMES ...]
Paths to files containing string identifiers for each
attributecolumn, being one file for each axis.
(default: None)
--Y Y If fitting the model to data, it represents the path
to a .csv file containing the known interaction matrix
between rows and columns data.If predicting, Y is the
destination path for the predicted values, formatted
as an interaction matrix in the same way described.
(default: None)
--path_model PATH_MODEL
When fitting: path to the location where the model
will be saved. When predicting: the saved model to be
used. (default: model.dict.pickle.gz)
--max_depth MAX_DEPTH
Maximum PBCT depth allowed. -1 will disable this
stopping criterion. (default: -1)
--min_samples_leaf MIN_SAMPLES_LEAF
Minimum number of sample pairs in the training set
required to arrive at each leaf. (default: 20)
--verbose, -v Show more detailed output (default: False)
--outdir OUTDIR Where to save results. (default: PBCT_results)
--k K [K ...], -k K [K ...]
Number of folds for cross-validation. (default: [3])
--diag Use independent TrTc sets for cross-validation, i.e.
with no overlapping rows or columns. (default: False)
--test_size TEST_SIZE [TEST_SIZE ...]
If between 0.0 and 1.0, represents the proportion of
the dataset to include in the TrTc split for each
axis, e.g.: .3 .5 means 30% of the rows and 50% of the
columns will be used as the TrTc set. If >= 1,
represents the absolute number of test samples in each
axis. If None, the values are set to the complements
of train_size. If a single value v is given, it will
be interpreted as (v, v). If train_size is also None,
it will be set to 0.25. (default: None)
--train_size TRAIN_SIZE [TRAIN_SIZE ...]
Same as test_size, but refers to the LrLc set
dimensions. (default: None)
--njobs NJOBS How many processes to spawn when cross-validating.
(default: None)
--random_state RANDOM_STATE
Random seed to use. (default: None)