What it does

A collection of Jupyter notebooks and Python scripts teaching RNN-based time series prediction with TensorFlow. Covers sine-wave forecasting with Gaussian processes and LSTMs, plus electricity price prediction using real 2015 energy market data. Each topic ships both standalone code and a narrated notebook.

The interesting bit

The repo includes slide decks from two Korean conference talks (TensorFlow-KR Meetup and KSC 2016), making it a small archaeological window into how early TensorFlow practitioners taught sequence modeling before Keras became the default frontend.

Key highlights

• Three worked examples: MNIST-with-RNN, sine-wave prediction (GP + LSTM), and electricity price forecasting
• Dual format: every example has both .py and .ipynb versions
• Real dataset included: 2015 electricity price history CSV, plus pointer to Energy Price Forecast 2016 competition data
• Adapted from an earlier tensorflow-lstm-regression project by Mourad Mourafiq

Caveats

• Pinned to TensorFlow r0.9 (released 2016); README notes deprecation warnings for tf.split_squeeze and tf.dnn that were already stale when written
• Author acknowledged update requests and targeted TensorFlow v1.2, but repo appears unmaintained since
• Dependencies (Python 3.4.4, pandas 0.16.2) are several years past end-of-life

Verdict

Worth a quick browse if you’re writing a history of TensorFlow pedagogy or need to understand legacy LSTM implementations. Everyone else should start with modern tf.keras or PyTorch tutorials — the concepts transfer, but the API archaeology required here is punishing.