HI team,

Please let me know how to interpret the below confision matrix and the query

Regards,

tony

Hi @train.bi

The shades of the block represents the number of values for the block. (You can also choose to display the values in the square box itself. that would be more convenient).

Consider the first block (0,0), looks like most of the images that were actually 0, were also predicted as 0.

Look at the block 2, 1… shows that some of the values which were actually 2, were predicted as 1.

Hi Aishwarya,

Thanks for your inputs.

confused with how do i figure out which has the least accuracy?

the answer is eight for how do we justify with the plot given

I would suggest plotting a mask´, showing the actual values over it.

Another point is the scale.

I assume this matrix does not come from any prediction probability or confidence?!

So, if it does just represent a given output metric (even if it is the nuber of hidden units voting for the particular class or spiking) you are probably better norming out the metric:

C’ = sum of your hidden outputs over the redundant hidden dimension / length of the redundant hidden dimension

If this shall represent a regression task, you should take more coefficients, representing your regression value, and manybe think about taking the logarithm instead.

0 was read as 0 and 6 was read as 6 largest number of time (diagnol). This is from the color scale. To decide out if 6 and 0 which is predicted best. We need actual number s. Atleast I can’t see the color difference between 0 and 6.