The "colorful" dimension

[ste (Stefano Giomo)]

I’ve just released the first version of the Colorful Dimension and the new Twin Peaks chart.

The Twin Peaks Chart

The Colorful Dimension Chart in two lines of code

data = ImageDataBunch.from_folder(untar_data(URLs.MNIST_SAMPLE),bs=1024)
# (1) Create custom ActivationsHistogram according to your needings
actsh = partial(ActivationsHistogram,modulesId=None,hMin=-10,hMax=10,nBins=200) 
# Add it to the callback_fns
learn = cnn_learner(data, models.resnet18, callback_fns=actsh, metrics=[accuracy])
# Fit: and see the Twin Peaks chart in action
learn.fit_one_cycle(4)
# (2) Customize and Plot the colorful chart!
learn.activations_histogram.plotActsHist(cols=20,figsize=(30,15),showEpochs=False)

You can find the article and the code here:

The Twin Peaks (Chart)

The Twin Peaks chart is a tool we can use to evaluate the health of our model in real time. It compares, for the last layer, the average of the activations histogram of the last quarter of batches with the activations histogram of the validation set for the same epoch. [twin_peaks_chart_live] TL;DR Seeing the Twin Peaks on activations histogram of the last layer is a good sign for a classifier. More separated peaks suggest better model “understanding” (accuracy and generalization). For a cla…