🚀 DLCore

DLCore is a lightweight, flexible deep learning framework built on PyTorch. It allows you to train and evaluate neural networks (RNN, GRU, LSTM, Transformer, CNN) easily with minimal boilerplate code. Perfect for regression and classification tasks.

✨ Features

• ✅ Supports RNN, GRU, LSTM, CNN, Transformer, Sequential
• ✅ Compatible with PyTorch tensors & NumPy arrays
• ✅ Built-in training loops, evaluation, metrics
• ✅ Early stopping & checkpoint saving
• ✅ Flexible metrics: MSE, MAE, R2, Accuracy, F1
• ✅ Lightweight and easy to integrate into existing projects

📦 Installation

You can install DLCore via pip (recommended) or from GitHub:

# Install from PyPI
pip install deeplcore

# Or install latest version from GitHub
git clone https://github.com/SOCIALPINE/dlcore.git
cd dlcore
pip install -r requirements.txt
pip install .

🏗 Quick Start

Import DLCore

import numpy as np
import os
import matplotlib.pyplot as plt
import dlcore as tk

# checkpoints
os.makedirs("checkpoints", exist_ok=True)

# ------------------------------
# 1) RNN
# ------------------------------
rnn_model = tk.RNN(input_size=10, hidden_size=20, num_layers=2)
trainer = tk.DLCore(
    rnn_model,
    epochs=20,
    lr=0.001,
    batch_size=32,
    metrics=['mse', 'mae', 'r2'],
    early_stopping=True,
    patience=3,
    checkpoint_path="checkpoints/rnn_best.pt",
    task='regression',
    verbose=1,
)

X_train = np.random.rand(200, 5, 10)
y_train = np.random.rand(200, 5, 20)
X_val = np.random.rand(50, 5, 10)
y_val = np.random.rand(50, 5, 20)

print("=== RNN Regression Training ===")
trainer.fit(X_train, y_train, validation_data=(X_val, y_val))
preds = trainer.predict(X_val)
print("Prediction shape:", preds.shape)

# ------------------------------
# 2) CNN
# ------------------------------
cnn_model = tk.CNN(in_channels=3, out_channels=16)
trainer_cnn = tk.DLCore(
    cnn_model,
    epochs=10,
    batch_size=32,
    metrics=['mse','mae','r2'],
    checkpoint_path="checkpoints/cnn_best.pt",
    task='regression',
    verbose=1,
)

X_cnn = np.random.rand(150, 3, 50)
y_cnn = np.random.rand(150, 16, 50)

print("\n=== CNN Regression Training ===")
trainer_cnn.fit(X_cnn, y_cnn)
preds_cnn = trainer_cnn.predict(X_cnn)
print("CNN Prediction shape:", preds_cnn.shape)

# ------------------------------
# 3) MLP
# ------------------------------
mlp_model = tk.Sequential(
    tk.Linear(20, 50),
    tk.ReLU(),
    tk.Linear(50, 3),
)

trainer_cls = tk.DLCore(
    mlp_model,
    epochs=30,
    lr=0.005,
    batch_size=64,
    loss_fn='cross_entropy',
    metrics=['accuracy', 'f1'],
    early_stopping=True,
    patience=5,
    checkpoint_path="checkpoints/mlp_cls_best.pt",
    task='classification',
    verbose=1,
)

X_cls_train = np.random.rand(500, 20)
y_cls_train = np.random.randint(0, 3, size=(500,))
X_cls_val = np.random.rand(100, 20)
y_cls_val = np.random.randint(0, 3, size=(100,))

print("\n=== MLP Classification Training ===")
trainer_cls.fit(X_cls_train, y_cls_train, validation_data=(X_cls_val, y_cls_val))
preds_cls = trainer_cls.predict(X_cls_val)
print("Classification Prediction shape:", preds_cls.shape)
print("Sample predictions:", preds_cls[:10])

# ------------------------------
# 4) GRU
# ------------------------------
gru_model = tk.GRU(input_size=10, hidden_size=15, num_layers=2)
trainer_gru = tk.DLCore(
    gru_model,
    epochs=15,
    batch_size=32,
    metrics=['mse', 'r2'],
    checkpoint_path="checkpoints/gru_best.pt",
    task='regression',
    verbose=1
)
X_gru = np.random.rand(100, 5, 10)
y_gru = np.random.rand(100, 5, 15)
trainer_gru.fit(X_gru, y_gru)
print("\nGRU Regression Prediction shape:", trainer_gru.predict(X_gru).shape)

# ------------------------------
# 5) Transformer
# ------------------------------
transformer_model = tk.TransformerEncoder(d_model=10, nhead=2, num_layers=2)
trainer_trans = tk.DLCore(
    transformer_model,
    epochs=10,
    batch_size=16,
    metrics=['mse','r2'],
    checkpoint_path="checkpoints/trans_best.pt",
    task='regression',
    verbose=1
)
X_trans = np.random.rand(50, 5, 10)  # (seq_len, batch, features)
y_trans = np.random.rand(50, 5, 10)
trainer_trans.fit(X_trans, y_trans)
print("\nTransformer Regression Prediction shape:", trainer_trans.predict(X_trans).shape)


trainer.save_model("checkpoints/rnn_saved.pt")

trainer.load_model("checkpoints/rnn_saved.pt")
print("RNN model save/load success!")

📊 Supported Metrics

🔧 Checkpoints & Early Stopping

DLCore supports:

trainer = DLCore(model=model, early_stopping=True, patience=5, checkpoint_path='best_model.pth')

• Saves the best model automatically
• Stops training when validation loss does not improve

💡 Examples

See the examples/ folder for ready-to-run examples:

• example.py

📄 License

MIT License © 2025 Feel free to use, modify, and distribute DLCore in your projects.

❤️ Support / Contribute

If you enjoy DLCore, give it a ⭐ and feel free to submit pull requests or issues!

Made with ❤️ by dev_pine