Full Training Loop
Intermediate
The problem, from first principles
The earlier classification questions define a score, turn it into a probability, and determine the direction to correct it. Training is the repeated process that connects those pieces so a classifier learns its parameters from all examples.
From theory to code
Implement train_logistic_regression by reusing linear, sigmoid, bce_gradient, and gd_step in their forward-to-update order.
Constraints
inputis(batch_size, in_features)and binarytargetis(batch_size,).- Return
(weight, bias)with shapes(1, in_features)and(1,). - Initialize both parameters to zero and reshape targets once to
(batch_size, 1). - Run exactly
epochsfull-batch updates; zero epochs returns the initialization. - Do not reimplement the imported helpers or use loops over samples.
Hints
Open one at a time. Each gives away a little more than the last.
Hint 1
The forward pass has two stages: a linear score followed by a probability activation.
Hint 2
Within each epoch, pass sigmoid(linear(...)) and the reshaped targets to bce_gradient, then hand its results to gd_step.
Keywords
classical-mlclassificationtraining-loop
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.