Page 1 of 1

Banbot V2

Posted: Fri Sep 14, 2018 6:33 pm
by OneBoss
Dear Gamemasters,

I take this opportunity to congratulate you for using learning machines in order to classify a message (string of words) as abusive/non-abusive.

At the same time, I suggest you use Convolutional neural networks or Random forests or Extreme Gradient boosting for classifications. Reasons :

1. Naive Bayes assumes the independent variables vector is identical and independent which is very rarely met in the real world

2. Naive bayes is computationally expensive as it calculates the conditional probability per Bayes theorem before classifying each row where as some of the deep learning techniques like CNNs/LSTMs are very easy to implement with a Tensor flow / Spark based implementation

In case you do decide to upgrade your algorithm and need a helping hand, Iā€™d love to volunteer as I specialize in NLP :}

Congratulations again!

Re: Banbot V2

Posted: Sun Sep 16, 2018 8:55 am
by [GM]Blade
OneBoss wrote: ā†‘
Fri Sep 14, 2018 6:33 pm
Dear Gamemasters,
2. Naive bayes is computationally expensive as it calculates the conditional probability per Bayes theorem before classifying each row where as some of the deep learning techniques like CNNs/LSTMs are very easy to implement with a Tensor flow / Spark based implementation
This is why we do deferred processing. It's useless to waste time on this kind of implementation that is so unproductive. We might as well consider such advanced AI/Neural Networks for boss behaviors.

Re: Banbot V2

Posted: Sun Sep 16, 2018 12:37 pm
by OneBoss
[GM]Blade wrote: ā†‘
Sun Sep 16, 2018 8:55 am
This is why we do deferred processing. It's useless to waste time on this kind of implementation that is so unproductive. We might as well consider such advanced AI/Neural Networks for boss behaviors.
Makes sense. AI for boss behaviour seems interesting!