LSTM TO STRING CATEGORICAL LABELS

Hi Please help with with this code, there are two questions,

1. Looks like my LSTM cannot achieve any better accuracy - what could be the cause?
2. At the very end of the code, I wanted to to plot a confusion chart, - 1 - I used a for loop to capture the predicted labels from the trained network, is there a one line command for this type of data structure?
3. Still on the confusion chart, what would be the best way to create the true labels set, I see the one I used in the function call for confusion chart is really incomplete, I am expecting the true labels set to be 40 x 5 matrix just like the test set.

[Ernest Modise - Kgamane]

Hi I managed to resolve the second part, I realized that I had not indexed my categorical lables properly on line 20 of the code.

I still want to know what the training can only achieve around 80 % accuracy. How can I improve this?

Answer 1

[Ernest Modise - Kgamane]

I realized that there was a problem in my data. I had some duplications, this has been sorted by cleaning my input file LSTMdataIn.xlsx

Training on single CPU.

|========================================================================================|

| Epoch | Iteration | Time Elapsed | Mini-batch | Mini-batch | Base Learning |

| | | (hh:mm:ss) | Accuracy | Loss | Rate |

|========================================================================================|

| 1 | 1 | 00:00:04 | 25.00% | 1.5810 | 0.5000 |

| 9 | 50 | 00:00:06 | 100.00% | 6.0340e-05 | 0.5000 |

| 10 | 60 | 00:00:07 | 100.00% | 4.7343e-05 | 0.5000 |

|========================================================================================|

Training finished: Max epochs completed.