def main():
# experiment settings
nruns = args.nruns if args.benchmark else 1
batch_sizes = [int(elem) for elem in args.batch]
# load data and instances to be explained
data = load_data()
predictor = load_model('assets/predictor.pkl') # download if not available locally
y_test, X_test_proc = data['all']['y']['test'], data['all']['X']['processed']['test']
logging.info(f"Test accuracy: {accuracy_score(y_test, predictor.predict(X_test_proc))}")
X_explain = data['all']['X']['processed']['test'].toarray() # instances to be explained
if args.workers == -1: # sequential benchmark
logging.info(f"Running sequential benchmark without ray ...")
distributed_opts = {'batch_size': None, 'n_cpus': None, 'actor_cpu_fraction': 1.0}
explainer = fit_kernel_shap_explainer(predictor, data, distributed_opts=distributed_opts)
run_explainer(explainer, X_explain, distributed_opts, nruns)
# run distributed benchmark or simply explain on a number of cores, depending on args.benchmark value
else:
workers_range = range(1, args.workers + 1) if args.benchmark == 1 else range(args.workers, args.workers + 1)
for workers in workers_range:
for batch_size in batch_sizes:
logging.info(f"Running experiment using {workers} actors...")
logging.info(f"Running experiment with batch size {batch_size}")
distributed_opts = {'batch_size': int(batch_size), 'n_cpus': workers, 'actor_cpu_fraction': 1.0}
explainer = fit_kernel_shap_explainer(predictor, data, distributed_opts)
run_explainer(explainer, X_explain, distributed_opts, nruns)
ray.shutdown()
def main():
# initialise ray
ray.init(address='auto')
# experiment settings
nruns = args.nruns
batch_sizes = [int(elem) for elem in args.batch]
# load data and instances to be explained
data = load_data()
predictor = load_model(
'assets/predictor.pkl') # download if not available locally
y_test, X_test_proc = data['all']['y']['test'], data['all']['X'][
'processed']['test']
logging.info(
f"Test accuracy: {accuracy_score(y_test, predictor.predict(X_test_proc))}"
)
X_explain = data['all']['X']['processed']['test'].toarray(
) # instances to be explained
distributed_opts = {'n_cpus': args.workers}
explainer = fit_kernel_shap_explainer(predictor, data, distributed_opts)
for batch_size in batch_sizes:
logging.info(f"Running experiment using {args.workers} actors...")
logging.info(f"Batch size: {batch_size}")
run_explainer(explainer, X_explain, distributed_opts, nruns,
batch_size)
def prepare_explainer_args(
data: Dict[str, Any]) -> Tuple[str, np.ndarray, dict, dict]:
"""
Extracts the name of the features (group_names) and the columns corresponding to each feature in the faeture matrix
(group_names) from the `data` dict and defines the explainer arguments. The background data necessary to initialise
the explainer is also extracted from the same dictionary.
Parameters
----------
data
A dictionary that contains all information necessary to initialise the explainer.
Returns
-------
A tuple containing the positional and keyword arguments necessary for initialising the explainers.
"""
groups = data['all']['groups']
group_names = data['all']['group_names']
background_data = data['background']['X']['preprocessed']
assert background_data.shape[0] == 100
init_kwargs = {'link': 'logit', 'feature_names': group_names, 'seed': 0}
fit_kwargs = {'groups': groups, 'group_names': group_names}
predictor = load_model(PREDICTOR_URL)
worker_args = (predictor, background_data, init_kwargs, fit_kwargs)
return worker_args
def main(path_to_X_explain, path_to_training_data):
nruns = args.nruns if args.benchmark else 1
batch_sizes = [int(elem) for elem in args.batch]
# load data and instances to be explained
predictor = load_model(
'assets/predictor.pkl') # download if not available locally
X_explain = read_file(path_to_X_explain)
data = np.array(read_file(path_to_training_data)).astype(np.float64)
X_explain = np.array(X_explain).astype(
np.float64) # instances to be explained
if args.workers == -1: # sequential benchmark
logging.info(f"Running sequential benchmark without ray ...")
distributed_opts = {
'batch_size': None,
'n_cpus': None,
'actor_cpu_fraction': 1.0
}
explainer = fit_kernel_shap_explainer(
predictor, data, distributed_opts=distributed_opts)
run_explainer(explainer, X_explain, distributed_opts, nruns)
# run distributed benchmark or simply explain on a number of cores, depending on args.benchmark value
else:
workers_range = range(1, args.workers +
1) if args.benchmark == 1 else range(
args.workers, args.workers + 1)
for workers in workers_range:
for batch_size in batch_sizes:
logging.info(f"Running experiment using {workers} actors...")
logging.info(
f"Running experiment with batch size {batch_size}")
distributed_opts = {
'batch_size': int(batch_size),
'n_cpus': workers,
'actor_cpu_fraction': 1.0
}
explainer = fit_kernel_shap_explainer(predictor, data,
distributed_opts)
run_explainer(explainer, X_explain, distributed_opts, nruns)
ray.shutdown()