torch-simpletrainer 0.1.1

A simple, flexible training framework for PyTorch models.

simpletrainer

🚀 Overview

simpletrainer keeps your PyTorch training tidy: metrics, early stopping, LR schedules, checkpoints, and optional W&B. Supports binary, multiclass, multilabel classification and regression.

---

🧩 Installation

Install name: torch-simpletrainer • Import name: simpletrainer

From PyPI

Install the correct PyTorch build for your platform first

pip install torch-simpletrainer

From source

git clone https://github.com/rahaahmadi/simpletrainer.git
cd simpletrainer
pip install -e .

---

⚙️ Quickstart Example

import torch
from torch import nn, optim
from torch.utils.data import DataLoader, TensorDataset
from simpletrainer import BaseTrainer

X = torch.randn(512, 16)
y = (torch.randn(512) > 0).float()
ds = TensorDataset(X, y)
train_loader = DataLoader(ds, batch_size=64, shuffle=True)
test_loader  = DataLoader(ds, batch_size=128)

model = nn.Sequential(nn.Linear(16, 32), nn.ReLU(), nn.Linear(32, 1))
criterion = nn.BCEWithLogitsLoss()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="max", patience=2)

trainer = BaseTrainer(
    model=model,
    train_loader=train_loader,
    test_loader=test_loader,
    criterion=criterion,
    optimizer=optimizer,
    scheduler=scheduler,
    task_type="binary",
    early_stopping_patience=5,
    grad_clip=1.0,
    wandb_project="my-project", # optional
)

best_epoch, best_metrics = trainer.fit(num_epochs=25, save_path="checkpoints/run/model")
print("Best epoch:", best_epoch)
print("Best metrics:", best_metrics)

---

✨ Features

• Tasks: binary classification, multiclass classification, multilabel classification, regression
• Metrics: AUC / F1 / Accuracy (classification), MSE / MAE / R² (regression)
• Training features: early stopping, gradient clipping, LR schedulers
• Checkpointing: save_model / load_model utilities
• Logging: optional Weights & Biases (auto-skips if not installed)

---

🤝 Contributing

Issues and PRs are welcome.