apply tSNE to the feature vectors output by a Convolutional Neural Network $($CNN$).1.$Load the cifar$10$data$.$Remember that some networks like the efficientNet expect unnormalized images. $2.$Build a CNN which accepts cifar$10$images as input. it makes sense to use GlobalAveragePooling$2$D after efficient net. If you use a network pre$-$trained on imagenet, make sure you need to reshape the data first

If you use efficientNet remember it accepts unnormalized data $--$ so don't divide by $255.3.$Feed the entire test data through the feature extraction part of the network. Use the predict function$/$method to get your output vectors. Use the predict function$/$method to get your output vectors. If your network is meant for classification, make sure not to take the final predictions $($class probabilities$).$ Take the input of the layer which would normally be fed into the MLP$.4.$Apply tSNE to the resulting vectors as usual. $5.$Visualize the tSNE output using a $2-$dimensional scatterplot, color point depending on the correct label. The labels$[$i$]$corresponds to the i$-$th class as specified by y$\_$train. I$\text{'}$d use both colors and labels. Use plt$.$legend$()$ to include the labels in the plot. What kind of classes$/$labels are most often confused? $6.$Visualize the tSNE output using a scatterplot, but this time show the image corresponding to each point. $7.$Use a network fine$-$tuned to cifar$10$and repeat the $2$tSNE plots.

train$/$fit the network on the train dataset. carefully unfreeze some of the layers