Product Category Classification Engine
This is a snapshot of the product category classification engine used by Two Tap. It uses glove word vectors and images to categorize a product to a unified taxonomy.
The current model is trained mostly on Apparel. It's really bad at electronics and other categories. There's an internal effort to improve this by adding new data.
How to use the it with the pre-trained weights.
- Clone the repo.
- Download the pretrained weights from here, unpack them, and place them in the models/ directory.
- You can edit things there.
The models/ directory should look like below:
How to train the model.
- Download training.csv from here and place it in the ml-data/ directory.
- Download the glove pre-trained vectors from here and place them in the ml-data/ directory.
- Run training/download_images.py to fetch all the images.
- Run training/train.py. This might take a long time.
Check the models/ directory for the model json and weights afterward.
The ml-data/ directory should look like below, but with more images:
Want access to the product images so you don't have to download them from the retailers? Email firstname.lastname@example.org.