def plot(model_name, dataset, datasize):
exp = "6"
path_manager = PathManager(model_name, dataset)
cpp = defaultdict(list)
nlci = defaultdict(list)
var_num = 0
for feature_type in AbsExp.get_attribute_method_names():
result_file = path_manager.result_json_path(exp, feature_type, size=datasize)
if os.path.isfile(result_file):
with open(result_file) as f:
results = json.loads(f.readlines()[0])
for value in results.values():
var_num = len(value["CPP"])
cpp[feature_type].append(value["CPP"])
nlci[feature_type].append(value["NLCI"])
pdf_name = path_manager.figure_path(datasize, exp)
pp = PdfPages(pdf_name)
values = []
names = []
# Filter condition
for name, instances in sorted(cpp.items(), key=Plot.sort_key, reverse=True):
value = [0] * (var_num + 1)
for instance in instances:
for idx, v in enumerate(instance):
value[idx + 1] += v
values.append([i / len(instances) for i in value])
names.append(name)
Plot.plot_line(values, names, pp, model_name, "Avg. CPP", "\#Changed Features", [0, 1.1], 0.5, (18, 6),
(20, 50))
values = []
names = []
for name, instances in sorted(nlci.items(), key=Plot.sort_key, reverse=True):
value = [0] * (var_num + 1)
for instance in instances:
for idx, v in enumerate(instance):
value[idx + 1] += v
values.append(value)
names.append(name)
logger.info("Number of instance {}".format(len(instances)))
Plot.plot_line(values, names, pp, model_name, "Avg. NLCI", "\#Changed Features", [0, 1100], 500, (18, 6),
(20, 50))
pp.close()
def plot(model_name, dataset, datasize):
exp = "7"
path_manager = PathManager(model_name, dataset)
cs = defaultdict(list)
for feature_type in AbsExp.get_attribute_method_names():
result_file = path_manager.result_json_path(
exp, feature_type, datasize)
if os.path.isfile(result_file):
with open(result_file) as f:
results = json.loads(f.readlines()[0])
for value in results.values():
cs[feature_type].append(value["Cosine"])
pdf_name = path_manager.figure_path(datasize, exp)
pp = PdfPages(pdf_name)
values, names = [], []
for i, j in sorted(cs.items(), key=Plot.sort_key, reverse=True):
values.append(sorted(j, reverse=True))
names.append(i)
Plot.plot_line(values, names, pp, model_name, "CS",
"Index of Instance", [-0.05, 1.1], 0.5, (18, 6),
(100, 300))
pp.close()
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--mdl", help="Model Name", choices=["LMT", "MLP"], required=True)
parser.add_argument("--dataset", help="data. E.g. FMNIST", choices=["FMNIST", "MNIST"], required=True)
parser.add_argument("--datasize", help="number of test images per class", required=True)
parser.add_argument("--explainer", help="name of explainer", choices=AbsExp.get_method_names())
parser.add_argument("--gpu", help="GPU used", choices=["cuda:1", "cuda:0", "cuda:2", "cuda:3"])
parser.add_argument("--task", help="GPU used", choices=["plot", "compute"], required=True)
parser.add_argument("--feature", help="Feature type", choices=AbsExp.get_attribute_method_names())
parsedArgs = parser.parse_args(sys.argv[1:])
model_name = parsedArgs.mdl
dataset = parsedArgs.dataset
datasize = int(parsedArgs.datasize)
task = parsedArgs.task
if task == "compute":
expln_name = parsedArgs.explainer
config.DEVICE = parsedArgs.gpu
feature_type = parsedArgs.feature
exp = Exp6(model_name, dataset, expln_name, datasize, feature_type)
exp.run()
else:
Exp6.plot(model_name, dataset, datasize)
logger.info("Finish")
