We have a [login to view URL] classification model trained and saved and load it and then run tests to see the predictions however the test predictions is giving very different results.
We have tested this by using the same data as was used for Training instead of new data and should get similar/same results from Testing but we are not seeing this.
We think there is an issue with either our scaling before we put in data or we are not reading the prediction properly. It would appear to be something simple but we are not familiar enough to see why.
We need a [login to view URL] Expert to look at the the file we use for training and the file we use for testing and fix the issue with the Prediction file giving inaccurate results.
We will consider this project completed once we can test the same data as the training data and see the similar/same accuracy.
Please note below is what our Test Prediction file creates after going through all the 13 different models we have. All of them give very different Test Prediction Accuracy compared do the saved Validation Accuracy we have from [login to view URL] from training.
The files are in a private Guthub repo, so please show interest in this job and I can send you an invite and you can look at the files and have access to the training data so you can test all this and see what is happening.
Many thanks in advance and good luck!
COMPLETED -> TESTING PREDICTION ON DATA
1 TEST ACCURACY -> 35 % TRAINING ACCURACY -> 84%
2 TEST ACCURACY -> 40 % TRAINING ACCURACY -> 81%
3 TEST ACCURACY -> 47 % TRAINING ACCURACY -> 82%
5 TEST ACCURACY -> 53 % TRAINING ACCURACY -> 82%
8 TEST ACCURACY -> 60 % TRAINING ACCURACY -> 83%
13 TEST ACCURACY -> 26 % TRAINING ACCURACY -> 94%
21 TEST ACCURACY -> 33 % TRAINING ACCURACY -> 93%
34 TEST ACCURACY -> 40 % TRAINING ACCURACY -> 88%
55 TEST ACCURACY -> 46 % TRAINING ACCURACY -> 90%
89 TEST ACCURACY -> 36 % TRAINING ACCURACY -> 95%
144 TEST ACCURACY -> 45 % TRAINING ACCURACY -> 95%
233 TEST ACCURACY -> 55 % TRAINING ACCURACY -> 95%
377 TEST ACCURACY -> 64 % TRAINING ACCURACY -> 96%
Hi, I can understand the bug you are facing, and to be honest this is a very typical problem most of ML projects face. I think the most probable cause would be some preprocessing not applied correctly while testing. J Plus
2 freelances font une offre moyenne de 125 $ pour ce travail
Hi, hope you are doing well. I am experienced data scientist with great passion for deep learning and web development. I have experience in working with Tensorflow JS. Recently i have worked on face recognition l Plus