prototorch_models/examples/glvq_spiral.py

77 lines
1.8 KiB
Python
Raw Normal View History

"""GLVQ example using the spiral dataset."""
2021-05-21 15:55:55 +00:00
import argparse
2021-06-04 20:21:28 +00:00
import prototorch as pt
import pytorch_lightning as pl
2021-05-21 16:54:47 +00:00
import torch
if __name__ == "__main__":
2021-05-21 15:55:55 +00:00
# Command-line arguments
parser = argparse.ArgumentParser()
parser = pl.Trainer.add_argparse_args(parser)
args = parser.parse_args()
# Dataset
2021-06-04 20:21:28 +00:00
train_ds = pt.datasets.Spiral(num_samples=500, noise=0.5)
# Dataloaders
2021-06-04 20:21:28 +00:00
train_loader = torch.utils.data.DataLoader(train_ds, batch_size=256)
# Hyperparameters
2021-05-25 13:41:10 +00:00
num_classes = 2
2021-06-04 20:21:28 +00:00
prototypes_per_class = 10
hparams = dict(
2021-05-25 13:41:10 +00:00
distribution=(num_classes, prototypes_per_class),
2021-06-04 20:21:28 +00:00
transfer_function="swish_beta",
2021-05-04 18:56:16 +00:00
transfer_beta=10.0,
2021-06-04 20:21:28 +00:00
proto_lr=0.1,
bb_lr=0.1,
input_dim=2,
latent_dim=2,
)
# Initialize the model
2021-06-04 20:21:28 +00:00
model = pt.models.GMLVQ(
hparams,
optimizer=torch.optim.Adam,
prototypes_initializer=pt.initializers.SSCI(train_ds, noise=1e-2),
2021-06-04 20:21:28 +00:00
)
# Callbacks
2021-06-04 20:21:28 +00:00
vis = pt.models.VisGLVQ2D(
train_ds,
show_last_only=False,
block=False,
)
pruning = pt.models.PruneLoserPrototypes(
threshold=0.01,
2021-06-04 20:21:28 +00:00
idle_epochs=10,
prune_quota_per_epoch=5,
frequency=5,
2021-06-04 20:21:28 +00:00
replace=True,
prototypes_initializer=pt.initializers.SSCI(train_ds, noise=1e-1),
2021-06-04 20:21:28 +00:00
verbose=True,
)
es = pl.callbacks.EarlyStopping(
monitor="train_loss",
min_delta=1.0,
patience=5,
mode="min",
check_on_train_epoch_end=True,
)
# Setup trainer
2021-05-21 15:55:55 +00:00
trainer = pl.Trainer.from_argparse_args(
args,
2021-06-04 20:21:28 +00:00
callbacks=[
vis,
es,
2021-06-04 20:21:28 +00:00
pruning,
],
2021-05-21 15:55:55 +00:00
terminate_on_nan=True,
)
# Training loop
trainer.fit(model, train_loader)