from cs231n.classifiers import Softmax

This Softmax doesn't contain predict and train functions in the code

I was not able to understand the significance of this statement. Any help is appreciable. 🍰

Assignment 1 Q2 Linear SVM
In assignment1/cs231n/classifiers/linear_svm.py
line 79: margins = np.maximum(0, scores - correct_class_scores +1)

It seems line 79 includes the same-class loss adding the total loss by a constant 1*num_train, which might be fine for training, but strictly speaking not correct.

the fundemental idea just like : (a - b)2 = a2 + b2 - 2ab

A = np.sum(np.square(X), axis=1) # 500 x 1 B = np.transpose(np.sum(np.square(X_train), axis=1)) # 1x5000 two_AB = 2*np.dot(X, self.X_train.T) # 500 x 5000 dicts = np.sqrt(A + B - two_AB)

how could it be broadcasted together with shapes (500,) (1,5000) ?

and anther one, at line 143,
for i in xrange(num_test):

range should be located here ?

cs231n/assignment1/two_layer_net.ipynb的Forward pass: compute scores中，correct_scores怎么来的，怎么能保证通过随机生成的矩阵经过运算后，能和corret_scores基本相同？

the fundemental idea just like : (a - b)2 = a2 + b2 - 2ab

A = np.sum(np.square(X), axis=1)   # 500 x 1

B = np.transpose(np.sum(np.square(X_train), axis=1))  # 1x5000

two_AB = 2*np.dot(X, self.X_train.T)   # 500 x 5000 

dicts = np.sqrt(A + B - two_AB)`

how could it be broadcasted together with shapes (500,) (1,5000) ?

what is meaning of f = lambda W: net.loss(X, y, reg=0.05)[0]? I am confused because I do not understand how net.loss act on different W?

In my opinion,
In line 266, the index of ar_cache should be self.num_layers-1 instead of self.num_layers due to the for loop you used in line 255(self.num_layers-1 is excluded here)
Therefore, a minor change should be made in line 294 as well.
Since you use the index of the output layer (self.num_layers) consistently, your code works perfectly. However, for numerical continuity, there should be self.num_layers-1.
Your code helps me a lot, Thanks so much!
Happy Spring Festival!

Hey, I've been using your code to compare against my own and you've done an amazing job helping out with going through the assignments with clean code.

In the dropout forward pass of assignment two I noticed you have
mask = (np.random.rand(*x.shape) >= p) / (1-p)
And this makes so that your drop out matches your p. But according to the class notes: # probability of keeping a unit active. higher = less dropout

so I believe the line should read:
mask = (np.random.rand(*x.shape) < p) / p

I'm curious because I hope others continue to use your code to learn and get through the class as I have 👍