Optimization callbacks and monitoring in SciPy - Step-by-Step Execution

Concept Flow - Optimization callbacks and monitoring

Start Optimization

↓

Iteration begins

↓

Evaluate function

↓

Call callback with current state

↓

Check stopping criteria

No Yes↓

Update variables

↓

Next iteration

The optimization runs in steps. After each step, it evaluates the function, calls the callback to monitor progress, then decides to continue or stop.

Execution Sample

SciPy

from scipy.optimize import minimize

def callback(xk):
    print(f"Current x: {xk}")

res = minimize(lambda x: (x-3)**2, x0=0, callback=callback)

This code runs an optimization to find x that minimizes (x-3)^2, printing x at each step.

Execution Table

Step	Current x	Function Value	Callback Output	Action
1	[0.0]	9.0	Current x: [0.0]	Update x towards 3
2	[1.8]	1.44	Current x: [1.8]	Update x towards 3
3	[2.52]	0.2304	Current x: [2.52]	Update x towards 3
4	[2.808]	0.0467	Current x: [2.808]	Update x towards 3
5	[2.965]	0.0012	Current x: [2.965]	Update x towards 3
6	[2.999]	0.0000	Current x: [2.999]	Converged, stop
7	[3.0]	0.0	Current x: [3.0]	Optimization ends

💡 Optimization stops when x is close enough to 3 and function value is near zero.

Variable Tracker

Variable	Start	After 1	After 2	After 3	After 4	After 5	After 6	Final
x	0.0	1.8	2.52	2.808	2.965	2.999	3.0	3.0
function_value	9.0	1.44	0.2304	0.0467	0.0012	0.0000	0.0	0.0

Key Moments - 2 Insights

Why does the callback print the current x before the variable updates?

The callback is called after the function evaluation at the current x but before the next update, as shown in execution_table rows where callback output matches current x before action.

Why does the optimization stop even though x is not exactly 3?

Optimization stops when the function value is close enough to zero (converged), not necessarily exactly at 3, as seen in step 6 where function value is near zero.

Visual Quiz - 3 Questions

Look at the execution table, what is the function value at step 3?

A 0.2304

B 1.44

C 0.0467

D 9.0

💡 Hint

Check the 'Function Value' column at step 3 in the execution_table.

At which step does the optimization decide to stop?

A Step 5

B Step 7

C Step 6

D Step 4

💡 Hint

Look at the 'Action' column where it says 'Converged, stop'.

If the callback was removed, what would change in the execution table?

A Function values would be different

B No callback output column and no printed current x values

C Optimization would not stop

D x values would not update

💡 Hint

Callback only monitors progress; it does not affect variable updates or stopping.

Concept Snapshot

Optimization callbacks let you monitor progress each step.
Callback runs after function evaluation but before variable update.
Use callback to print or log current variables.
Optimization stops when criteria met, not necessarily exact solution.
Callbacks do not change optimization behavior, only observe it.

Full Transcript

In optimization with scipy, the process starts by evaluating the function at an initial guess. After each evaluation, a callback function is called with the current variables. This callback can print or log the current state to monitor progress. Then the optimizer updates the variables to move closer to the minimum. This repeats until the stopping criteria are met, such as the function value being close enough to zero. The callback helps track the optimization without changing its behavior. The example shows x moving from 0 to 3, with the callback printing x at each step. The optimization stops when x is near 3 and the function value is near zero.

Practice

1. What is the main purpose of using a callback function during optimization in scipy.optimize?

easy

A. To monitor and control the optimization process step-by-step

B. To automatically fix errors in the optimization function

C. To speed up the optimization by parallel processing

D. To save the final result to a file

Solution

1. Step 1: Understand the role of callbacks in optimization

   Callbacks are functions called at each iteration to observe or influence the optimization process.

2. Step 2: Identify the main use of callbacks

   They allow monitoring progress, logging, or stopping optimization early based on conditions.

3. Final Answer:

   To monitor and control the optimization process step-by-step -> Option A

4. Quick Check:

   Callbacks = monitor/control optimization [OK]

Hint: Callbacks watch optimization progress stepwise [OK]

Common Mistakes:

• Thinking callbacks fix errors automatically
• Assuming callbacks speed up optimization
• Believing callbacks save results directly

2. Which of the following is the correct way to define a callback function for scipy.optimize.minimize that prints the current parameter values at each iteration?

easy

A. def callback(): print(f"Current params: {xk}")

B. def callback(xk): return xk

C. def callback(xk, fk): print(f"Current params: {fk}")

D. def callback(xk): print(f"Current params: {xk}")

Solution

1. Step 1: Check callback function signature for scipy.optimize.minimize

   The callback receives one argument: the current parameter vector xk.

2. Step 2: Verify the function prints current parameters correctly

   def callback(xk): print(f"Current params: {xk}") defines callback(xk) and prints xk, which is correct.

3. Final Answer:

   def callback(xk): print(f"Current params: {xk}") -> Option D

4. Quick Check:

   Callback signature = one argument (xk) [OK]

Hint: Callback gets current params as single argument [OK]

Common Mistakes:

• Defining callback without parameters
• Using wrong parameter names or extra parameters
• Returning values instead of printing

3. Given the following code snippet, what will be printed during the optimization?

from scipy.optimize import minimize

def callback(xk):
    print(f"Step: {xk[0]:.2f}, {xk[1]:.2f}")

def func(x):
    return (x[0]-1)**2 + (x[1]-2)**2

res = minimize(func, [0, 0], callback=callback)

medium

A. Multiple lines showing parameter values approaching (1.00, 2.00)

B. Only one line showing initial parameters [0.00, 0.00]

C. No output because callback is ignored

D. Error because callback function has wrong signature

Solution

1. Step 1: Understand the callback usage in minimize

   The callback prints the current parameters at each iteration during optimization.

2. Step 2: Analyze the expected output

   Since the function minimizes distance to (1,2), parameters will update stepwise, printing multiple lines approaching (1.00, 2.00).

3. Final Answer:

   Multiple lines showing parameter values approaching (1.00, 2.00) -> Option A

4. Quick Check:

   Callback prints each step params [OK]

Hint: Callback prints each iteration's parameters [OK]

Common Mistakes:

• Assuming callback prints only once
• Thinking callback is ignored by minimize
• Believing callback signature is incorrect here

4. You wrote this callback to stop optimization early when the first parameter exceeds 0.5:

def callback(xk):
    if xk[0] > 0.5:
        return True

But the optimization does not stop early. What is the likely problem?

medium

A. The callback must raise an exception to stop optimization

B. The callback function must return False to stop optimization

C. Returning True does not stop optimization in scipy.optimize.minimize

D. The callback must be passed as a keyword argument named 'stop_callback'

Solution

1. Step 1: Understand callback behavior in scipy.optimize.minimize

   Callbacks can monitor progress but returning True does not stop the optimization.

2. Step 2: Identify correct way to stop optimization early

   To stop early, you must raise an exception or use other control mechanisms; returning True is ignored.

3. Final Answer:

   Returning True does not stop optimization in scipy.optimize.minimize -> Option C

4. Quick Check:

   Return True ≠ stop optimization [OK]

Hint: Returning True in callback won't stop optimization [OK]

Common Mistakes:

• Expecting return True to stop optimization
• Using wrong callback argument name
• Not raising exception to stop optimization

5. You want to log the optimization progress to a list and stop optimization if the function value goes below 0.01. Which callback implementation correctly achieves this with scipy.optimize.minimize?

hard

A. log = []
def callback(xk):
val = (xk[0]-1)**2 + (xk[1]-2)**2
log.append(val)
if val < 0.01:
return True

res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback)

B. log = []
def callback(xk):
val = (xk[0]-1)**2 + (xk[1]-2)**2
log.append(val)
if val < 0.01:
raise StopIteration

res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback)

C. log = []
def callback(xk, fk):
log.append(fk)
if fk < 0.01:
raise StopIteration

res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback)

D. log = []
def callback(xk):
val = (xk[0]-1)**2 + (xk[1]-2)**2
log.append(val)
if val < 0.01:
return False

res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback)

Solution

1. Step 1: Understand callback signature and logging

   The callback receives current parameters xk. We compute function value manually and append to log.

2. Step 2: Stopping optimization early

   Returning True or False does not stop optimization; raising an exception like StopIteration is a valid way.

3. Step 3: Verify options

   log = [] def callback(xk): val = (xk[0]-1)**2 + (xk[1]-2)**2 log.append(val) if val < 0.01: raise StopIteration res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback) correctly logs values and raises StopIteration when value < 0.01. log = [] def callback(xk): val = (xk[0]-1)**2 + (xk[1]-2)**2 log.append(val) if val < 0.01: return True res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback) returns True (ignored). log = [] def callback(xk, fk): log.append(fk) if fk < 0.01: raise StopIteration res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback) has wrong callback signature (two args). log = [] def callback(xk): val = (xk[0]-1)**2 + (xk[1]-2)**2 log.append(val) if val < 0.01: return False res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback) returns False (ignored).

4. Final Answer:

   log = [] def callback(xk): val = (xk[0]-1)**2 + (xk[1]-2)**2 log.append(val) if val < 0.01: raise StopIteration res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0,0], callback=callback) -> Option B

5. Quick Check:

   Raise exception to stop + log values [OK]

Hint: Raise exception in callback to stop optimization [OK]

Common Mistakes:

• Returning True or False expecting to stop optimization
• Using wrong callback signature with two arguments
• Not computing function value inside callback