My website will use a Neural Network to predict thing based on user data. The user can select the data to be used in training the network and then use their trained network to predict things.
I am using a framework to create, train and query the networks. This uses Java. The framework has persistence for saving a network to an XML file.
What is the best way to store these files? I can see several potential ideas, but I need help on choosing which is best:
- Save each network to a separate XML file with a name that is stored in the database. Load this each time.
- Save all the networks to the same XML file with each network having a different name that is stored in the database.
- Somehow pass what would normally be written to an XML file to the Django site for writing to the database. This would need to be returned to the Java code when a prediction needs to be made.
I am able to do 1 or 2, but I think their performance will be quite limited and I am on shared hosting at the moment, so I don't know how pleased they would be with thousands of files. Also, after adding a few thousand records to one XML file, I was noticing a massive performance hit on saving to it.
If I were able to implement version 3 somehow I think it would be best. No issues with separate processes accessing the database and I think performance would be better. Not to mention having no files lying around.
However, the stuff in the neural network framework I am using (Encog) for saving to a file needs access to a Java file object, not a string that could be saved to a database. Unless there is some Java magic I can do here (I know very little Java), the only way I can see of doing this would be with a temporary files but I don't know if this is the correct way to do it.
I would appreciate any ideas on the best way to implement any of the above 3 ideas or any alternatives. Thanks!
Update: After talking with my Boss, we both agr开发者_C百科eed that idea 3 is the best. It seems like the 'proper' way to do it. I hope someone can help us on this one. Thanks.
I think I have the best solution and as no one has answered it is what I will go with for the moment.
I am creating a temporary file in Java and using this for saving the data. Then before exiting, I read that file and output it to stdout so that the python code can pick it up and store it in the database. Thankfully, it is always a similar size so I can limit the number of characters on the database field to 3500.
If anyone gets a better solution, please comment to tell me, but for the moment, this is what I will use.
精彩评论