将 keras 模型保存到数据库

Question

I created a keras model (tensorflow) and want to store it in my MS SQL Server database. What is the best way to do that? pyodbc.Binary(model) throws an error. I would prefer a way without storing the model in the file system first.

Thanks for any help

Answer 1

It seems that there is no clean solution to directly store a model incl. weights into the database. I decided to store the model as h5 file in the filesystem and upload it from there into the database as a backup. For predictions I load anyway the model from the filesystem as it is much faster than getting it from the database for each prediction.

Answer 2

The best approach would be to save it as a file in the system and just save the path in the database. This technique is usually used to store large files like images since databases usually struggle with them.

Answer 3

Well, you can save a model as a JSON like:

json_string = model.to_json()

And then cast it to string and save normally to the BD.

Or use native MS SQL solution for JSON fields explained here .