def run_fn(args, evaluate_explanation=True):
np.random.seed()
dataset = args[0]
trial = args[1]
depth = args[2]
size = args[3]
rate = args[4]
reg = args[5]
name = args2name(dataset, trial, depth, size, rate, reg)
cwd = os.getcwd()
os.makedirs(name)
os.chdir(name)
manager = "regression"
source = DATASET_PATH + dataset + ".csv"
shape = [size] * depth
out = eval(manager,
source,
hidden_layer_sizes=shape,
learning_rate=rate,
regularizer="Causal1D",
c=reg,
stddev_reg=0.5,
evaluate_explanation=evaluate_explanation,
stop_on_loss=True)
with open("out.json", "w") as f:
json.dump(out, f)
os.chdir(cwd)
def run_fn(trial, type, evaluate_explanation = True):
np.random.seed()
dataset = "msd"
depth = 5
size = 100
rate = 0.001
reg = 0.1
name = args2name(dataset, trial, depth, size, rate, reg)
cwd = os.getcwd()
os.makedirs(name)
os.chdir(name)
manager = "msd"
source = DATASET_PATH
shape = [size] * depth
out, epoch = eval(manager, source,
hidden_layer_sizes = shape,
learning_rate = rate,
regularizer = type, c = reg, stddev_reg = 0.5,
evaluate_explanation = False,
stop_on_loss = True,
min_epochs = 10, stopping_epochs = 10)
with open("out.json", "w") as f:
json.dump(out, f)
os.chdir(cwd)
return epoch
def run_fn(args, evaluate_explanation = True):
np.random.seed()
dataset = args[0]
trial = args[1]
depth = args[2]
size = args[3]
rate = args[4]
name = args2name(dataset, trial, depth, size, rate)
cwd = os.getcwd()
os.makedirs(name)
os.chdir(name)
manager = "msd"
source = DATASET_PATH
shape = [size] * depth
out = eval(manager, source,
hidden_layer_sizes = shape,
learning_rate = rate,
evaluate_explanation = evaluate_explanation,
stop_on_loss = True,
min_epochs = 10, stopping_epochs = 10)
with open("out.json", "w") as f:
json.dump(out, f)
os.chdir(cwd)
p.map(run_fn, configs)
p.close()
p.join()
if flag_agg:
agg = {}
config = {}
list_means = {}
for reg in regs:
mean_acc = 0.0
mean_lime = 0.0
for trial in trials:
name = args2name(dataset, trial, depth, size, rate, reg)
with open(name + "/out.json") as f:
results = json.load(f)
mean_acc += results["test_acc"]
mean_lime += results["lime_causal_metric"][0]
mean_acc /= len(trials)
mean_lime /= len(trials)
list_means[str(reg)] = [mean_acc, mean_lime]
with open("results.json", "w") as f:
json.dump(list_means, f)
if flag_run_baselines:
def baseline(name):