Hello,
Here I wanted to discuss about some much better results that I got with Weka compared to Tensorflow. ¿Would be it possible to replicate exactly the Weka MLP model I used in Tensorflow?
I wanted to discriminate between three letters of de ASL (American Sign Language): A, B and C. The three converted to the Media Pipe Holistic and each instance with only one frame. I had 60 instances (20 for each letter) with the pose, face, left hand and right hand (x, y, z and visibility). When there were no values I put a 0.
With Tensorflow I got bad results so I tried with Weka. I got more than a 90% precision with a simple Multi Layer Perceptron with only 5 neurons in the hidden layer and sigmoid. I tried with 3 different seeds with similar results. With the Tensorflow that proposed ChatGPT to replicate Weka model I got a maximum of 75%. I used a 80/20 split in Weka.
Here I do not know how to attach the csv file I used.
It would be possible with Tensorflow to have a model that replicates exactly this MLP from Weka?
Thankyou very much
Carles