def plot1(name, xlist, ylist):
pylab.grid(True)
for i in xrange(len(ylist)):
plt.plot(xlist, ylist[i], colors[i % len(colors)], marker=markers[i % len(markers)])
plt.legend(label, loc='center right')
plt.xlim(xlist[0] / 2, xlist[-1] + xlist[0] / 2)
plt.xlabel('Node Set Percentage')
plt.ylabel('Locality Percentage')
plt.title('{}: step={}'.format(name, len(xlist)))
fname = name + '_exp1.png'
plt.savefig(utils.get_result_fname(fname))
plt.clf()
def plot2(name, labels, xlist, ylist):
pylab.grid(True)
for i in xrange(len(ylist)):
plt.plot(xlist, ylist[i], colors[i % len(colors)], marker=markers[i % len(markers)])
plt.legend(labels, loc='center right')
plt.xlim(xlist[0] / 2, xlist[-1] + xlist[0] / 2)
plt.xlabel('Number of Parts')
plt.ylabel('Locality Percentage')
plt.title('{}: Average Localities for different Parts'.format(name))
fname = name + '_exp2.png'
plt.savefig(utils.get_result_fname(fname))
plt.clf()
def plot1_bar(name, xlist, ylist):
pylab.grid(True)
width = 0.15
ind = np.arange(len(xlist))
fig, ax = plt.subplots()
for i in xrange(len(ylist)):
ax.bar(ind + i * width, ylist[i], width, color=colors[i % 3])
ax.set_xticks(ind + width)
ax.set_xticklabels(xlist)
ax.legend(label, loc='center')
plt.title('{}: step={}'.format(name, len(xlist)))
fname = name + '_exp1_bar.png'
plt.savefig(utils.get_result_fname(fname))
plt.clf()
def plot2_random_partition(name, xlist, ylist):
yy = []
for x, y in zip(xlist, ylist):
yy.append(x * y)
plt.plot(xlist, yy)
plt.xlabel(r'$\textsf{num}$')
plt.ylabel(r'$\textsf{l}*\textsf{num}$')
plt.title('{}: Random Partition Locality'.format(name))
plt.legend(['Parts*Locality'], loc='best')
plt.ylim([0.5, 1.5])
pylab.grid(True)
fname = name + '_exp2_rnd.png'
plt.savefig(utils.get_result_fname(fname))
plt.clf()
train_dl, val_dl = load_data(config.dataset,
config.batch_size,
num_workers=0)
print(config.key2val, config.max_epochs)
# Train model
trainer = pl.Trainer(max_epochs=config.max_epochs,
logger=logger,
gpus=utils.get_n_gpus())
trainer.fit(fodvae, train_dl, val_dl)
fodvae_best_version = fodvae
# Save best version
save_path = str(MODELS_DIR / model_fname(config))
torch.save(fodvae_best_version.state_dict(), save_path)
return evaluate(config, fodvae_best_version, logger, return_results)
if __name__ == "__main__":
args = parse_args()
config = utils.Config(args)
return_results = config.experiment == "ablative"
results = main(config, return_results=return_results)
print(results)
if return_results:
with open(RESULTS_DIR / utils.get_result_fname(config), "w") as f:
f.write(json.dumps(results, indent=2))
print(
f"Written results to {(RESULTS_DIR / utils.get_result_fname(args)).relative_to(PROJECT_DIR)}"
)
