Timeline for Keras: Prediction performance does not match accuracy
Current License: CC BY-SA 4.0
14 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 23, 2020 at 8:46 | answer | added | malelis | timeline score: 0 | |
| Mar 23, 2020 at 1:33 | answer | added | Artimizia Dias | timeline score: 2 | |
| Aug 30, 2018 at 2:25 | comment | added | Aditya | But we should avoid -1...( It means auto detection) Maybe works sometimes or it might break sometimes.. | |
| Aug 30, 2018 at 0:06 | answer | added | Rob Campbell | timeline score: 1 | |
| Aug 29, 2018 at 20:36 | comment | added | Rob Campbell | And I have verified that the classes are correct, if I look at test_generator.filenames and test_generator.classes, the classes match up with the directory names. | |
| Aug 29, 2018 at 20:34 | comment | added | Rob Campbell | Hi: I divided by 10K because that's how many test images there are. I found that in a prior example, but don't recall which one or have a link handy. I also took the -1 on the axis from another example. It was initially 1, but both seem to give the same results. | |
| Aug 29, 2018 at 17:48 | comment | added | Aditya | Also they say why dividing by 10k? WE Need to divide by the #test set images passed on a batch | |
| Aug 29, 2018 at 17:32 | comment | added | Aditya | Hey After asking them, they say that you should verify your label classes...(high probability that it's messed up) Like class 0 is 1 or something like that, also why a -1 on the axis??? | |
| Aug 29, 2018 at 17:27 | comment | added | Aditya | Hey thanks for your response, I am also excited to know the Reason behind this, so please do t forget if you figured it out... Had asked few of my friends, they are looking into it! | |
| Aug 29, 2018 at 17:21 | comment | added | Rob Campbell | I'm less worried about model performance at this point, the problem is that the predictions of the test set are clearly wrong. I adapted your model to the mnist_png files and ran it, it worked great, final epoch was loss: 0.0293 - acc: 0.9911. The test set had good results too (loss: 0.0137 - acc: 0.9952), but when I checked the accuracy from the results produced by model.predict_generator (i.e. the last blockquote in my original question) the accuracy was 0.09. I suspect it's the indexing, maybe I'll take this over to the Keras forum. | |
| Aug 22, 2018 at 1:22 | comment | added | Aditya | Maybe you should verify the directory structure with this datascience.stackexchange.com/a/34281/35644, Also try visualising the CNN Layers to see why it's so less | |
| Aug 22, 2018 at 1:17 | comment | added | Aditya | I am not sure why the loss is so less, but here is my kernel which achieves .995 on MNIST, Also maybe dense layers needs to be tweaked as directly jumping from 512-->10 doesn't seem a good idea to me! kaggle.com/adityaecdrid/mnist-with-keras-for-beginners-99457 | |
| Aug 21, 2018 at 20:15 | review | First posts | |||
| Aug 21, 2018 at 22:22 | |||||
| Aug 21, 2018 at 20:13 | history | asked | Rob Campbell | CC BY-SA 4.0 |