Weird unexpected behavior

[jochemstoel]

When I run your example 02_hardware-software.js I sometimes (not always) get weird unpredictable characters from the training data input prepended to the output. I did not modify the script or the training data.

When I run the file I get stuff stuff like: Category: plyhardware Category: ore memoryhardware Category: soma pwae minghardware Category: nit test failedsoftware

I don't understand this. Do you?

Node 11.5.0 Windows 10

[birksy89]

Very interesting - I got the same thing on the 3rd run of 02_hardware-software.js

Category: ofted a lystop-inputstart-outputhardware

Node 11.2.0 Windows 10

[Praneet460]

I think the reason behind is that somehow the LSTM model is also training itself with the text in the category section. To solve this issue we need to represent our categories with the help of the dummy variables like replace software with [0] and replace hardware with [1] After replacing and re-training your model you will find it working fine.