Final_Project_MachineLearning_in_TensorFlow_Berkeley

This is the final project of Berkeley extension's Machine Learning Course in TensorFlow.

Project Proposal:

The dataset, obtained from Kaggle is a list of ~20,000 recipes listed by rating, nutritional information and assigned category that is already parsed. I plan to predict the public rating of recipes based on continuous and categorical features given in the dataset.

The feature size is large (678 count) but extremely sparse. Hence, I believe that the dataset would make a great candidate to build a model combining wide linear model and a deep feed-forward neural network (using DNNLinearCombinedClassifier). The wide linear model is able to memorize interactions with data but not able to generalize learned interactions on new data. The deep model generalizes well but is unable to learn exceptions within the data. It is intended that the wide and deep model combines the two models and is able to generalize while learning exceptions (https://arxiv.org/abs/1606.07792).

The code is in Jupyter notebook. There are two notebooks:

1. Model to predict if food is a dessert: Final_project_DNNClassifier_predict_dessert.ipynb
2. Model to predict rating: Final_project_DNNClassifier_predict_rating.ipynb

Instructions for running the model:

1. Download and unzip Epicurious dataset: epi_r.csv. Save it in a folder and use path to refer to in notebook
2. Run packages and functions in the notebook.
3. The model can run three types of model: "wide", "deep" and "wide + deep".
4. Define model_type and model_dir and run test_model_accuracy(model_type, model_dir)
5. Go to your define model_dir in terminal and run Tensorboard: tensorboard --logdir ./