I am so confused about the objective of this Lab? Would greatly appreciate if someone could please provide detailed expaination to my following questions?
Q1: What are we trying to predict here? Is 4 deferent classifications (0,1,2,3?)
Q2: X_train is (2000 by 2) why the output is 2 by 4?
Q3: in the content below says "the outputs are not probabilities, but can range from large negative numbers to large positive numbers. " Is not -log(ez/sum(ez) should be greater than 0 hence, the outputs should be range from 0 to large positive numbers?
Q4: in the content below says: " To select the most likely category, the softmax is not required. One can find the index of the largest output using [np.argmax()]. What this means? does it mean we can just use .argmax() to make prediction instead of using .softmax?(numpy.argmax — NumPy v1.26 Manual)."
A1 The training dataset consists of 2000 examples, where each example has two coordinates, and the values are clustered around four different coordinate pairs. centers = [[-5, 2], [-2, -2], [1, 2], [5, -2]]
These clusters are numbered 0, 1, 2, and 3 by the ‘y_train’ values.
However, since we want to create a model that gives the best classification for each example, we are doing a classification method, rather than directly trying to predict the integer values 0 through 3 (which would be a linear regression method).
This is why the model is compiled to use SparseCategoricalCrossentropy and either softmax, or “from_logits = True”. Both of these methods will automatically convert the y_train values into one-hot outputs (where there are four outputs, one for each classification), as seen in the model definition here:
Thank you Tom for your detailed responses. much appreciated.
I could not see a response for Q4, could you please provide an answer to it as well?
Q4: in the content below says: " To select the most likely category, the softmax is not required. One can find the index of the largest output using [np.argmax()]. I don’t get what this means? does it mean we can just use .argmax() to make prediction instead of using .softmax?(numpy.argmax — NumPy v1.26 Manual)."
