Model parameters inspection in PyTorch - Model Metrics & Evaluation

Metrics & Evaluation - Model parameters inspection

Which metric matters for Model parameters inspection and WHY

When inspecting model parameters, the key metrics are parameter count and parameter distribution. These tell us how complex the model is and if parameters are learning useful patterns. We look at parameter values to check if they are too large, too small, or stuck at zero, which can indicate problems like underfitting or overfitting.

Confusion matrix or equivalent visualization

Model parameters inspection does not use a confusion matrix. Instead, we use parameter histograms or summary tables showing counts and statistics like mean and standard deviation of weights and biases.

Parameter summary example:

Layer: Conv2d(3, 16, kernel=3x3)
- Total params: 432
- Mean weight: 0.012
- Std weight: 0.045
- Min weight: -0.12
- Max weight: 0.15

Layer: Linear(128, 10)
- Total params: 1290
- Mean weight: 0.005
- Std weight: 0.03
- Min weight: -0.08
- Max weight: 0.09
    

Tradeoff: Model size vs performance

A model with too few parameters may not learn well (underfitting). Too many parameters can memorize training data but fail on new data (overfitting). Inspecting parameters helps find a balance. For example, a small model with 1,000 parameters might be fast but less accurate, while a large model with 1 million parameters might be accurate but slow and prone to overfitting.

What "good" vs "bad" parameter inspection looks like

Good: Parameters have varied values, not all zeros or very large numbers. Weight distributions look normal (bell-shaped). Parameter count matches model design.

Bad: Parameters are all zeros or very close to zero (dead model). Parameters have extremely large values (exploding weights). Parameter count is unexpectedly low or high, indicating a coding error.

Common pitfalls in parameter inspection

• Ignoring parameter scale: Very large or very small weights can cause training issues.
• Confusing parameter count with model quality: More parameters do not always mean better models.
• Not checking biases separately: Biases affect output shifts and need inspection.
• Overlooking frozen parameters: Parameters not updating during training can cause poor learning.

Self-check question

Your model has 10,000 parameters, but after training, all weights in one layer are zero. Is this good? Why or why not?

Answer: This is not good. All zero weights mean that layer is not learning anything. It could be a bug or a problem with training. You should investigate why weights are stuck at zero.

Key Result

Inspecting parameter counts and distributions helps detect model complexity issues and training problems early.

Practice

1. What does the named_parameters() method in a PyTorch model return?

easy

A. The optimizer's learning rate

B. An iterator over tuples of parameter names and their tensors

C. The model's training loss values

D. A list of all model layers

Solution

1. Step 1: Understand the purpose of named_parameters()

   This method is designed to give access to each parameter's name and its tensor, which includes weights and biases.

2. Step 2: Identify what it returns

   It returns an iterator over tuples, where each tuple contains the parameter's name and the tensor itself.

3. Final Answer:

   An iterator over tuples of parameter names and their tensors -> Option B

4. Quick Check:

   named_parameters() = iterator of (name, tensor) [OK]

Hint: Remember: named_parameters() lists names and tensors [OK]

Common Mistakes:

• Confusing parameters with layers
• Thinking it returns loss or optimizer info
• Assuming it returns just tensors without names

2. Which of the following code snippets correctly prints the names and shapes of all parameters in a PyTorch model model?

easy

A. for name, param in model.named_parameters(): print(name, param.shape)

B. for param in model.parameters(): print(param.name, param.shape)

C. print(model.parameters().names())

D. model.print_parameters()

Solution

1. Step 1: Check correct method usage

   The method named_parameters() returns (name, parameter) pairs, so unpacking into two variables is correct.

2. Step 2: Verify printing name and shape

   Accessing param.shape gives the tensor shape. for name, param in model.named_parameters(): print(name, param.shape) correctly prints both.

3. Final Answer:

   for name, param in model.named_parameters(): print(name, param.shape) -> Option A

4. Quick Check:

   Use named_parameters() with unpacking [OK]

Hint: Use named_parameters() to get names and tensors [OK]

Common Mistakes:

• Trying to access .name on param tensor (no such attribute)
• Using parameters() without names
• Calling non-existent methods like print_parameters()

3. Given the following PyTorch model code, what will be the output of the print statement?

import torch.nn as nn

class SimpleNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc1 = nn.Linear(4, 3)
        self.fc2 = nn.Linear(3, 2)

model = SimpleNet()
params = list(model.named_parameters())
print(len(params))

medium

A. 2

B. 6

C. 4

D. 3

Solution

1. Step 1: Count parameters in each layer

   Each nn.Linear has weights and biases, so 2 parameters per layer.

2. Step 2: Calculate total parameters

   Two layers x 2 parameters each = 4 parameters total.

3. Final Answer:

   4 -> Option C

4. Quick Check:

   2 layers x 2 params = 4 [OK]

Hint: Each Linear layer has weight and bias parameters [OK]

Common Mistakes:

• Counting only layers, not parameters
• Assuming one parameter per layer
• Confusing number of parameters with tensor sizes

4. You wrote this code to print model parameter names and shapes but get an error:

for name, param in model.parameters():
    print(name, param.shape)

What is the error and how to fix it?

medium

A. The loop syntax is wrong; use for param, name in model.parameters()

B. param.shape is invalid; use param.size() instead

C. You must call model.parameters().items() to get name and param

D. model.parameters() returns tensors only; use model.named_parameters() to get names

Solution

1. Step 1: Understand what model.parameters() returns

   This method returns only parameter tensors, no names.

2. Step 2: Correct method to get names and parameters

   Use model.named_parameters() which returns (name, param) pairs.

3. Final Answer:

   model.parameters() returns tensors only; use model.named_parameters() to get names -> Option D

4. Quick Check:

   parameters() = tensors only, named_parameters() = (name, tensor) [OK]

Hint: Use named_parameters() to get names with tensors [OK]

Common Mistakes:

• Trying to unpack parameters() into name and param
• Using .items() on parameters() which is not a dict
• Swapping variable order in for loop

5. You want to calculate the total number of trainable parameters in a PyTorch model model. Which code snippet correctly computes this?

hard

A. sum(p.numel() for p in model.parameters() if p.requires_grad)

B. len(model.named_parameters())

C. sum(p.shape for p in model.parameters())

D. model.count_parameters()

Solution

1. Step 1: Understand what counts as trainable parameters

   Only parameters with requires_grad=True are trainable.

2. Step 2: Use numel() to count elements in each tensor

   Summing p.numel() for all trainable parameters gives total count.

3. Final Answer:

   sum(p.numel() for p in model.parameters() if p.requires_grad) -> Option A

4. Quick Check:

   Sum numel() of trainable params = total count [OK]

Hint: Sum numel() of parameters with requires_grad True [OK]

Common Mistakes:

• Using len() which counts parameters, not elements
• Trying to sum shapes directly (shapes are tuples)
• Assuming model has count_parameters() method