Logging parameters and metrics in MLOps - Step-by-Step Execution

Process Flow - Logging parameters and metrics

Start Training

↓

Set Parameters

↓

Log Parameters

↓

Train Model

↓

Evaluate Model

↓

Log Metrics

↓

End Training

This flow shows how parameters are set and logged before training, then metrics are logged after evaluation.

Execution Sample

MLOps

params = {"lr": 0.01, "epochs": 5}
log_params(params)
metrics = {"accuracy": 0.85, "loss": 0.35}
log_metrics(metrics)

This code logs training parameters first, then logs evaluation metrics after training.

Process Table

Step	Action	Input	Logged Data	Output/State
1	Set parameters	{"lr": 0.01, "epochs": 5}	None	Parameters ready for training
2	Log parameters	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	Parameters saved in log
3	Train model	Parameters	None	Model trained with given parameters
4	Evaluate model	Trained model	None	Metrics calculated
5	Log metrics	{"accuracy": 0.85, "loss": 0.35}	{"accuracy": 0.85, "loss": 0.35}	Metrics saved in log

💡 All parameters and metrics logged after training and evaluation complete

Status Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	After Step 5
params	{}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}
metrics	{}	{}	{}	{}	{"accuracy": 0.85, "loss": 0.35}	{"accuracy": 0.85, "loss": 0.35}
log_storage	{}	{}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5}	{"lr": 0.01, "epochs": 5, "accuracy": 0.85, "loss": 0.35}

Key Moments - 3 Insights

Why do we log parameters before training instead of after?

Logging parameters before training ensures the exact settings used are saved, as shown in step 2 of the execution_table where parameters are logged right after being set.

Can metrics be logged before training is complete?

No, metrics depend on model evaluation after training, so they are logged only after evaluation as shown in step 5 of the execution_table.

What happens if parameters or metrics are not logged?

Without logging, you lose track of what settings were used or how well the model performed, making it hard to reproduce or improve results, as implied by the log_storage variable in variable_tracker.

Visual Quiz - 3 Questions

Look at the execution_table, what data is logged at step 2?

A Training parameters

B Training metrics

C Model weights

D Evaluation results

💡 Hint

Check the 'Logged Data' column at step 2 in the execution_table

At which step does the metrics variable first get a value?

A Step 1

B Step 3

C Step 4

D Step 5

💡 Hint

Look at the variable_tracker row for 'metrics' and see when it changes from empty to having values

If we skip logging parameters, what changes in the variable_tracker?

A params variable would be empty

B log_storage would not contain parameters after step 2

C metrics would not be logged

D Model training would fail

💡 Hint

Refer to the 'log_storage' row in variable_tracker and step 2 in execution_table

Concept Snapshot

Logging parameters and metrics:
- Set parameters before training
- Log parameters immediately after setting
- Train and evaluate model
- Log metrics after evaluation
- Logging ensures reproducibility and tracking

Full Transcript

This visual execution shows how parameters and metrics are logged in a machine learning workflow. First, parameters like learning rate and epochs are set and logged before training starts. Then the model trains using these parameters. After training, the model is evaluated to produce metrics such as accuracy and loss. These metrics are then logged. The execution table traces each step, showing when parameters and metrics are saved. The variable tracker shows how variables change over time. Key moments clarify why logging order matters. The quiz tests understanding of when and what data is logged. This process helps keep track of experiments and results clearly.

Practice

1.

What is the main purpose of logging parameters in machine learning experiments?

easy

A. To record the settings used during model training

B. To measure the model's accuracy on test data

C. To save the final trained model file

D. To visualize the model's predictions

Solution

1. Step 1: Understand what parameters are

   Parameters are the settings or configurations used to train a model, like learning rate or number of layers.

2. Step 2: Identify the purpose of logging parameters

   Logging parameters helps keep track of these settings so you can compare different training runs.

3. Final Answer:

   To record the settings used during model training -> Option A

4. Quick Check:

   Logging parameters = record training settings [OK]

Hint: Parameters = training settings, metrics = performance [OK]

Common Mistakes:

• Confusing parameters with metrics
• Thinking logging saves the model file
• Assuming logging is for visualization

2.

Which of the following is the correct way to log a metric named accuracy with value 0.95 using a typical MLOps logging function log_metric?

easy

A. log_metric('accuracy', 0.95)

B. log_metric(accuracy=0.95)

C. log_metric('accuracy': 0.95)

D. log_metric(0.95, 'accuracy')

Solution

1. Step 1: Understand typical function syntax

   Logging functions usually take the metric name as a string first, then the value as a number.

2. Step 2: Check each option's syntax

   log_metric('accuracy', 0.95) uses correct syntax: function name, string key, numeric value. log_metric(accuracy=0.95) uses keyword argument which may not be supported. log_metric('accuracy': 0.95) uses invalid syntax with colon inside parentheses. log_metric(0.95, 'accuracy') reverses arguments incorrectly.

3. Final Answer:

   log_metric('accuracy', 0.95) -> Option A

4. Quick Check:

   Function(metric_name, value) = correct syntax [OK]

Hint: Metric name first as string, then value [OK]

Common Mistakes:

• Using colon instead of comma in function call
• Passing arguments in wrong order
• Using keyword arguments when not supported

3.

Given the following code snippet, what will be the output logged for the metric loss?

log_metric('loss', 0.25)
log_metric('loss', 0.20)
log_metric('loss', 0.15)

medium

A. Only the last value 0.15 is logged for 'loss'

B. An error occurs because 'loss' is logged multiple times

C. All three values 0.25, 0.20, and 0.15 are logged separately

D. The first value 0.25 overwrites the others

Solution

1. Step 1: Understand metric logging behavior

   Most MLOps tools allow logging multiple values for the same metric over time to track progress.

2. Step 2: Analyze the code snippet

   The code logs 'loss' three times with different values. Each call records a new metric value, not overwriting previous ones.

3. Final Answer:

   All three values 0.25, 0.20, and 0.15 are logged separately -> Option C

4. Quick Check:

   Multiple logs for same metric = multiple entries [OK]

Hint: Repeated metric logs add entries, not overwrite [OK]

Common Mistakes:

• Assuming repeated logs overwrite previous values
• Expecting an error on duplicate metric names
• Thinking only one value per metric is allowed

4.

Identify the error in this code snippet for logging a parameter batch_size with value 32:

log_param(batch_size, '32')

medium

A. Function name should be log_metric instead of log_param

B. Value should be a number, not a string

C. No error, the code is correct

D. Parameter name should be a string, not a variable

Solution

1. Step 1: Check parameter name argument

   The parameter name must be a string literal like 'batch_size', not a bare variable name.

2. Step 2: Check value argument

   Value can be string or number depending on context; '32' as string is acceptable here.

3. Final Answer:

   Parameter name should be a string, not a variable -> Option D

4. Quick Check:

   Parameter name = string literal [OK]

Hint: Parameter names must be quoted strings [OK]

Common Mistakes:

• Passing parameter name without quotes
• Confusing log_param with log_metric
• Thinking value must always be numeric

5.

You want to log both parameters and metrics for a training run using the following code:

log_param('learning_rate', 0.01)
log_param('optimizer', 'adam')
log_metric('accuracy', 0.92)
log_metric('loss', 0.1)

Which of these statements is true about the logged data?

hard

A. Metrics record model settings; parameters record model performance

B. Parameters record model settings; metrics record model performance

C. Both parameters and metrics record model performance

D. Both parameters and metrics record model settings

Solution

1. Step 1: Understand the role of parameters

   Parameters like learning rate and optimizer are settings used to train the model.

2. Step 2: Understand the role of metrics

   Metrics like accuracy and loss measure how well the model performs after training.

3. Final Answer:

   Parameters record model settings; metrics record model performance -> Option B

4. Quick Check:

   Parameters = settings, Metrics = performance [OK]

Hint: Parameters = settings, metrics = results [OK]

Common Mistakes:

• Mixing up parameters and metrics roles
• Thinking metrics are settings
• Assuming parameters measure performance