Random seed management in MLOps - Step-by-Step Execution

Process Flow - Random seed management

Set random seed

↓

Initialize random generator

↓

Run stochastic process

↓

Get reproducible output

↓

Use output for training/testing

↓

Repeat with same seed -> same results

↓

Change seed -> different results

This flow shows how setting a random seed fixes the starting point of randomness, making results repeatable across runs.

Execution Sample

MLOps

import random
random.seed(42)
print(random.randint(1, 10))
print(random.randint(1, 10))

This code sets a random seed and prints two random numbers, which will be the same every time it runs.

Process Table

Step	Action	Seed State	Random Number Generated	Output
1	Set seed to 42	Seed=42	N/A	No output
2	Generate first random int	Seed=42 (updated internally)	Random int between 1-10	2
3	Generate second random int	Seed=updated	Random int between 1-10	1
4	End of execution	Seed=updated	N/A	Outputs: 2, 1

💡 Random numbers generated based on seed 42, ensuring reproducibility.

Status Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	Final
seed	None	42	internal updated	internal updated	internal updated
random_number_1	None	None	2	2	2
random_number_2	None	None	None	1	1

Key Moments - 3 Insights

Why do the random numbers stay the same every time we run the code?

Because setting the seed (Step 1) fixes the starting point of the random number generator, so the sequence of numbers generated (Steps 2 and 3) is always the same.

What happens if we don't set a seed before generating random numbers?

Without setting a seed, the random number generator starts from a different state each time, so the numbers will differ on each run (not shown in table but implied).

Does the seed variable itself change after generating numbers?

The seed is used to initialize the generator; internally the generator state updates after each number, but the original seed value remains conceptually fixed (see 'internal updated' in variable_tracker).

Visual Quiz - 3 Questions

Look at the execution table, what is the first random number generated at Step 2?

A 2

B 1

C 42

D None

💡 Hint

Check the 'Random Number Generated' and 'Output' columns at Step 2 in the execution_table.

At which step does the seed get set to 42?

A Step 2

B Step 1

C Step 3

D Step 4

💡 Hint

Look at the 'Action' column in the execution_table to find when the seed is set.

If we change the seed to 100, what will happen to the output numbers?

A They will be the same as with seed 42

B No numbers will be generated

C They will be different from seed 42

D The seed will reset to 42 automatically

💡 Hint

Refer to the concept_flow where changing the seed leads to different results.

Concept Snapshot

Random seed management:
- Use random.seed(value) to fix randomness
- Fixing seed makes outputs reproducible
- Same seed = same random sequence
- Different seed = different sequence
- Useful for debugging and consistent experiments

Full Transcript

Random seed management means setting a starting point for random number generation so results repeat every time. We set the seed using random.seed(42). Then, when we generate random numbers, they come out the same on every run. This helps in machine learning experiments to get consistent results. The seed initializes the random generator's state. Each call to generate a random number updates this state internally. If we change the seed, the sequence of random numbers changes. Without setting a seed, the numbers vary each run. This trace showed setting seed 42 and generating two random numbers: 2 and 1, always the same on every run.

Practice

1. What is the main purpose of setting a random seed in machine learning experiments?

easy

A. To make the results reproducible and consistent across runs

B. To speed up the training process

C. To increase the randomness of the model

D. To reduce the size of the dataset

Solution

1. Step 1: Understand the role of randomness in experiments

   Randomness affects initialization and data shuffling, causing different results each run.

2. Step 2: Identify the effect of setting a seed

   Setting a seed fixes randomness so results are the same every time.

3. Final Answer:

   To make the results reproducible and consistent across runs -> Option A

4. Quick Check:

   Random seed = reproducibility [OK]

Hint: Random seed fixes randomness for repeatable results [OK]

Common Mistakes:

• Thinking seed speeds up training
• Believing seed increases randomness
• Confusing seed with dataset size

2. Which of the following Python code snippets correctly sets the random seed for both Python's random and NumPy libraries?

easy

A. import random
import numpy as np
random.seed(42)
np.seed(42)

B. import random
import numpy as np
random.seed(42)
np.random.seed(42)

C. import random
import numpy as np
random.seed = 42
np.random.seed = 42

D. import random
import numpy as np
random.set_seed(42)
np.set_seed(42)

Solution

1. Step 1: Recall correct seed setting methods

   Python's random uses random.seed(value), NumPy uses np.random.seed(value).

2. Step 2: Check each option's syntax

   import random import numpy as np random.seed(42) np.random.seed(42) uses correct functions. Others use non-existent set_seed, incorrect assignments to seed, or np.seed(42) which doesn't exist.

3. Final Answer:

   import random import numpy as np random.seed(42) np.random.seed(42) -> Option B

4. Quick Check:

   random.seed() and np.random.seed() are correct [OK]

Hint: Use .seed() method, not .set_seed or assignment [OK]

Common Mistakes:

• Using random.set_seed instead of random.seed
• Assigning seed as a variable instead of calling method
• Calling np.seed instead of np.random.seed

3. Consider the following Python code snippet:

import random
random.seed(123)
print([random.randint(1, 10) for _ in range(3)])
random.seed(123)
print([random.randint(1, 10) for _ in range(3)])

What will be the output?

medium

A. [[3, 2, 7], [4, 5, 6]]

B. [[1, 10, 2], [1, 10, 2]]

C. [[3, 2, 7], [3, 2, 7]]

D. [[1, 10, 2], [4, 5, 6]]

Solution

1. Step 1: Understand effect of setting seed before generating numbers

   Setting seed resets the random number generator to a fixed state.

2. Step 2: Predict output of two identical seed calls

   Both lists will be identical because the seed is reset before each list generation.

3. Final Answer:

   [3, 2, 7], [3, 2, 7] -> Option C

4. Quick Check:

   Same seed = same random sequence [OK]

Hint: Resetting seed repeats the same random sequence [OK]

Common Mistakes:

• Assuming different outputs after resetting seed
• Confusing seed effect with random state progression
• Ignoring that seed resets generator state

4. You have the following code snippet that aims to fix randomness but still produces different results each run:

import random
random.seed(42)
print(random.randint(1, 100))
import numpy as np
np.random.seed(42)
print(np.random.randint(1, 100))

What is the most likely reason for the non-reproducible results?

medium

A. The seed is set only for Python random and NumPy separately, but another library uses randomness

B. The random seed is set after generating random numbers

C. The seed value 42 is too small to fix randomness

D. The print statements cause randomness to reset

Solution

1. Step 1: Analyze seed setting for Python random and NumPy

   Seeds are set correctly for both libraries before generating numbers.

2. Step 2: Consider other sources of randomness

   If another library (e.g., TensorFlow, PyTorch) uses randomness but seed is not set there, results vary.

3. Final Answer:

   Seed set only for Python random and NumPy, but another library uses randomness -> Option A

4. Quick Check:

   All libraries need seed set for full reproducibility [OK]

Hint: Set seed in all libraries that use randomness [OK]

Common Mistakes:

• Thinking seed value size matters
• Believing print affects randomness
• Assuming seed order is wrong here

5. You want to ensure full reproducibility of a machine learning experiment using Python's random, NumPy, and PyTorch. Which of the following code snippets correctly sets seeds for all three libraries and disables nondeterministic behavior in PyTorch?

hard

A. import random
import numpy as np
import torch
random.seed(123)
np.random.seed(123)
torch.manual_seed(123)

B. import random
import numpy as np
import torch
random.seed(123)
np.random.seed(123)
torch.manual_seed(123)
torch.set_deterministic(True)

C. import random
import numpy as np
import torch
random.seed(123)
np.random.seed(123)
torch.manual_seed(123)
torch.deterministic = True

D. import random
import numpy as np
import torch
random.seed(123)
np.random.seed(123)
torch.manual_seed(123)
torch.use_deterministic_algorithms(True)

Solution

1. Step 1: Set seeds for Python random, NumPy, and PyTorch

   Use random.seed(), np.random.seed(), and torch.manual_seed() with the same value.

2. Step 2: Enable deterministic algorithms in PyTorch

   Use torch.use_deterministic_algorithms(True) to disable nondeterministic ops.

3. Final Answer:

   import random import numpy as np import torch random.seed(123) np.random.seed(123) torch.manual_seed(123) torch.use_deterministic_algorithms(True) -> Option D

4. Quick Check:

   All seeds set + deterministic mode = full reproducibility [OK]

Hint: Set all seeds and enable deterministic mode in PyTorch [OK]

Common Mistakes:

• Using non-existent torch.set_deterministic method
• Assigning torch.deterministic instead of calling function
• Forgetting to enable deterministic algorithms in PyTorch