38 lines
1.0 KiB
Python
38 lines
1.0 KiB
Python
"""k-NN example using the Iris dataset."""
|
|
|
|
import prototorch as pt
|
|
import pytorch_lightning as pl
|
|
import torch
|
|
|
|
if __name__ == "__main__":
|
|
# Dataset
|
|
from sklearn.datasets import load_iris
|
|
x_train, y_train = load_iris(return_X_y=True)
|
|
x_train = x_train[:, [0, 2]]
|
|
train_ds = pt.datasets.NumpyDataset(x_train, y_train)
|
|
|
|
# Dataloaders
|
|
train_loader = torch.utils.data.DataLoader(train_ds,
|
|
num_workers=0,
|
|
batch_size=150)
|
|
|
|
# Hyperparameters
|
|
hparams = dict(k=20)
|
|
|
|
# Initialize the model
|
|
model = pt.models.KNN(hparams, data=train_ds)
|
|
|
|
# Callbacks
|
|
vis = pt.models.VisGLVQ2D(data=(x_train, y_train), resolution=200)
|
|
|
|
# Setup trainer
|
|
trainer = pl.Trainer(max_epochs=1, callbacks=[vis], gpus=0)
|
|
|
|
# Training loop
|
|
# This is only for visualization. k-NN has no training phase.
|
|
trainer.fit(model, train_loader)
|
|
|
|
# Recall
|
|
y_pred = model.predict(torch.tensor(x_train))
|
|
print(y_pred)
|