experiment/main.py

80 lines
2.2 KiB
Python
Raw Permalink Normal View History

"""Write a short description of your experiment here."""
import logging
import lightning as L
import torch
from torch.utils.data import DataLoader, random_split
from protothor.callbacks.normalization import OmegaTraceNormalization
from protothor.data.preprocessing import Standardizer
from lightning.pytorch.callbacks import EarlyStopping
from protothor.data.toy import Iris
from protothor.functional.initialization import zero_initialization
from protothor.lightning.metric_module import MetricModule
from protothor.models.gmlvq import generate_model as generate_gmlvq
from protothor.nn.collectable_loss import EvaluateExposedLosses
from protothor.nn.container import find_instances
logging.basicConfig(level=logging.INFO)
def main():
# 1 - Get Dataset
data = Iris([0, 2])
train_ds, val_ds = random_split(data, [0.7, 0.3])
standardize = Standardizer(train_ds)
# 2 - Create Dataloaders
train_loader = DataLoader(
standardize(train_ds),
shuffle=True,
batch_size=len(train_ds),
)
val_loader = DataLoader(
standardize(val_ds),
shuffle=False,
batch_size=len(val_ds),
)
# 3 - Initialize Prototypes
labels, positions = zero_initialization(data)
# 4 - Generate Torch Model
model = generate_gmlvq(positions, labels)
omega_layer = find_instances(model, torch.nn.Linear)[0]
# 5 - Initialize Lightning Module
module = MetricModule(model, EvaluateExposedLosses(model))
# 6 - Define Callbacks
callbacks = [
OmegaTraceNormalization(omega_layer),
]
callbacks.append(
EarlyStopping(
monitor="accuracy/validation",
min_delta=0.0002,
patience=50,
mode="max",
verbose=False,
check_on_train_epoch_end=True,
)
)
# 7 - Define Trainer
trainer = L.Trainer(detect_anomaly=True, max_epochs=1000, callbacks=callbacks)
# 8 - Train Model
trainer.fit(module, train_loader, val_loader)
# 9 - Analyse results
omega = omega_layer.weight.detach()
classification_correlation_matrix = omega.T @ omega
logging.info(classification_correlation_matrix)
if __name__ == "__main__":
main()