I have thousands of CSV files containing 75 rows / 6 colums of integer data. Each file is paired with one of ~10 fixed strings and an array of 2 integers.
I want to see if is possible to train a model in Tensorflow JS to accurately match the CSV files with one of ~10 fixed short length strings or separately with a small 1D array containing 2 values. (separate model training process expected) I need someone experienced with Tensorflow JS who can quickly determine the best configuration for this scenario.
I am proficient with web development but am completely new to Tensorflow JS and short on time so i would like one initial [login to view URL] file with a JS inline script which loads the JS files from a local directory, parses the CSV data.
The files will be defined in an array with filename, string and nested array of 2 integers.
[["filename", 'elephant', [2, 0],
There are ~ 10 possible strings all short. If it matters there is fixed 15 different possible combinations of two integers.
I would like basic advice included on things like:
- how to increase training time,
- how the model could possibly be exported and progress saved in the future
- what is the best layers and algrorithms, loss method etc to be used in this scenario and what other options can be tried if the model fails to make accurate predictions.
- how to scale down the amount of training data if necessary due to hardware limitations
Accurate predictions is obviously not a requirement!