IS-Sentence-Completion

The goal of this project is to implement a sentence completion mechanism based on a recurrent neural network in Python and evaluate the quality and problems of this approach.

Project structure

📦IS-Sentence-Completion
┣ 📂doc
┃ ┣ 📂baseline
┃ ┣ 📂datasets
┃ ┣ 📂models
┃ ┣ 📂optimization
┃ ┗ 📂tokenizer
┣ 📂logs
┃ ┗ 📂training
┣ 📂res
┃ ┣ 📂datasets
┃ ┃ ┗ 📜...
┃ ┣ 📂models
┃ ┃ ┣ 📜...
┃ ┃ ┗ 📜readme.txt
┃ ┗ 📂tokenizers
┃ ┗ 📜...
┣ 📜is_autocomplete.ipynb
┗ 📜README.md

The resources and diagrams used in the documentation can be found in the doc folder. The logs/training folder contains TensorBoard logs (including network graph and metric data) for visualization. Datasets, models and corresponding tokenizers are saved in the res folder and the code as well as markdown documentation can be found in is_autocomplete.ipynb.

Installation / How to run

First open a Terminal / Powershell window in the project folder.

1. Install Jupyter Notebook:
   • Linux & Windows: pip install notebook
2. Optional: Installing dependencies manually:
   • Linux & Windows: pip install -r requirements.txt
3. Optional: Tensorflow-directml for training on GPU:
   • Linux & Windows: pip install tensorflow-directml
4. Run the notebook
   • Linux: jupyter notebook
   • Windows: jupyter-notebook.exe .