KaggleNotify 🔔

Get Telegram notifications for your Kaggle training runs — epoch updates, crash alerts, and completion messages. Zero config. One line of code.

---

✨ Features

Feature	Description
📊 Epoch updates	Loss, accuracy + trend arrows ↑↓ after every epoch
🏆 Best model alert	Fires whenever val_accuracy hits a new high
⏱ ETA	Time elapsed shown in every message
❌ Crash alerts	Full traceback + CUDA OOM detection sent to Telegram
🔇 Fail-silent	Telegram errors never crash your training
🧩 4 frameworks	Keras, PyTorch, HuggingFace, Sklearn

---

🚀 Quick Start

from kaggle_notify import setup, KerasNotifyCallback

notifier = setup("My Experiment")  # sends a test ping immediately

model.fit(X_train, y_train,
          callbacks=[KerasNotifyCallback(notifier)])

---

📦 Installation on Kaggle

# In a Kaggle notebook cell:
!pip install requests  # already installed, but just in case
!wget https://raw.githubusercontent.com/XMGMAX/kaggle-notify/main/kaggle_notify.py

---

🔑 Setup (one-time)

See SETUP.md for the full BotFather + Kaggle Secrets walkthrough.

TL;DR:

1. Create a bot via @BotFather → get BOT_TOKEN
2. Get your Chat ID from @userinfobot
3. Add both as Kaggle Secrets: TELEGRAM_BOT_TOKEN and TELEGRAM_CHAT_ID

---

📬 Example Telegram Messages

Epoch update:

📊 My Experiment — Epoch 5/10  ⏱ 12m 34s
  loss: 0.3421 ↓
  accuracy: 0.8812 ↑
  val_loss: 0.4102 ↓
  val_accuracy: 0.8654 ↑

Completion:

🎉 Training Complete!
📌 My Experiment
⏱ Time: 47m 32s
📈 Best Val Accuracy: 0.8921

Crash alert:

❌ Training Crashed!
⏱ Failed at: 23m 11s
💥 RuntimeError: CUDA out of memory
💡 Tip: Reduce batch size or use gradient checkpointing.
📋 Traceback...

---

🧩 Framework Examples

Framework	File
Keras / TensorFlow	examples/keras_example.py
PyTorch	examples/pytorch_example.py
HuggingFace	examples/huggingface_example.py
Scikit-learn	examples/sklearn_example.py

---

📄 License

MIT — Built by XMGMAX