Understanding the output

[dhgoratela]

I need help in understanding the output of the model. My questiions are as follows:

1. The model output is stored in a variable logits which is a n*3 matrix where n is the number of sentences. What does each column signify in the output matrix?

2. For sentiment_score variable, we calculate the difference between the fisrt and second column of model's output matrix. What is the intution behind calculating this difference?

3. I see that predictions variable is always evaluating None. What is the intution behing calculating that variable value?

[dhgoratela]

Let me know if the understanding is correct. for the first question:

1. Every column of model's output matrix is respectively an output class (positive [col1], negative [col2] and neutral [col3])

[doguaraci]

Let me know if the understanding is correct. for the first question:

1. Every column of model's output matrix is respectively an output class (positive [col1], negative [col2] and neutral [col3])

That is correct.

2. The sentiment score is one way to interpret the outcome. Subtract negative softmax value from the positive. We use it to transform the score into a continuous range from -1 to 1.

3. Could you explain a bit more what you mean by this?

[dhgoratela]

Sure..
In case I provide only one sentence as input, the logits variable evaluates a 1x3 array; Following that we do an argmax which gives a 1x1 array with any one of following value in it (0, 1, 2). Then when we do np.squeeze in this 1x1 array to calculate predictions variable, it evaluates nothing..
Pardon me, cannot share a screenshot. Let me know if I could explain you the issue..