Comparing experiment runs in MLOps - Step-by-Step Execution

Process Flow - Comparing experiment runs

Start: Select experiments

↓

Fetch run data

↓

Extract metrics and parameters

↓

Align runs for comparison

↓

Visualize differences

↓

Analyze results and decide

This flow shows how experiment runs are selected, their data fetched and aligned, then compared visually to analyze differences.

Execution Sample

MLOps

runs = fetch_runs(['run1', 'run2'])
metrics = extract_metrics(runs)
compare(metrics)
visualize_comparison()

This code fetches two experiment runs, extracts their metrics, compares them, and visualizes the differences.

Process Table

Step	Action	Input	Output	Notes
1	Select experiment runs	['run1', 'run2']	Runs selected	User chooses runs to compare
2	Fetch run data	Runs selected	Raw data for run1 and run2	Data includes params, metrics, tags
3	Extract metrics	Raw data	Metrics dict for each run	Focus on key performance metrics
4	Align runs	Metrics dicts	Aligned metrics table	Ensures metrics correspond across runs
5	Visualize comparison	Aligned metrics	Comparison chart/table	Shows differences clearly
6	Analyze results	Comparison chart	Insights on performance	User decides best run
7	Exit	N/A	Comparison complete	Process ends

💡 All selected runs compared and visualized for analysis

Status Tracker

Variable	Start	After Step 2	After Step 3	After Step 4	Final
runs	[]	['run1', 'run2']	['run1', 'run2']	['run1', 'run2']	['run1', 'run2']
raw_data	{}	{'run1': {...}, 'run2': {...}}	{'run1': {...}, 'run2': {...}}	{'run1': {...}, 'run2': {...}}	{'run1': {...}, 'run2': {...}}
metrics	{}	{}	{'run1': {'accuracy': 0.9, 'loss': 0.1}, 'run2': {'accuracy': 0.85, 'loss': 0.15}}	{'accuracy': [0.9, 0.85], 'loss': [0.1, 0.15]}	{'accuracy': [0.9, 0.85], 'loss': [0.1, 0.15]}
comparison_chart	null	null	null	null	Rendered chart/table

Key Moments - 3 Insights

Why do we need to align metrics before comparing runs?

Because different runs might have different sets of metrics or order. Aligning ensures we compare the same metrics side by side, as shown in step 4 of the execution_table.

What happens if a metric is missing in one run?

The alignment step handles missing metrics by marking them as absent or null, so the comparison chart can show gaps or differences clearly, preventing misleading results.

Why is visualization important after comparison?

Visualization helps quickly spot differences and trends between runs, making it easier to analyze results and decide which run performed better, as seen in step 5.

Visual Quiz - 3 Questions

Look at the execution_table, at which step are the metrics aligned for comparison?

A Step 5

B Step 3

C Step 4

D Step 2

💡 Hint

Check the 'Action' column in execution_table row with 'Align runs'

According to variable_tracker, what is the value of 'metrics' after Step 3?

A Empty dictionary {}

B {'run1': {'accuracy': 0.9, 'loss': 0.1}, 'run2': {'accuracy': 0.85, 'loss': 0.15}}

C Raw data for runs

D Rendered chart/table

💡 Hint

Look at the 'metrics' row under 'After Step 3' column in variable_tracker

If a new run 'run3' is added, which step would need to be repeated to include it in the comparison?

A All steps from 1 to 5

B Steps 1 and 2

C Step 1 only

D Steps 3 and 4

💡 Hint

Adding a run affects selection, fetching data, extracting metrics, aligning, and visualization

Concept Snapshot

Comparing experiment runs:
1. Select runs to compare
2. Fetch their data (params, metrics)
3. Extract key metrics
4. Align metrics across runs
5. Visualize differences
6. Analyze to choose best run

Full Transcript

This visual execution shows how to compare experiment runs step-by-step. First, runs are selected. Then their raw data is fetched, including parameters and metrics. Next, key metrics are extracted from each run. These metrics are aligned so that the same metrics from different runs are side by side. After alignment, a visualization like a chart or table is created to show differences clearly. Finally, the user analyzes the visualization to decide which run performed better. Variables like 'runs', 'raw_data', and 'metrics' change as the process moves forward. Key moments include understanding why alignment is needed, handling missing metrics, and the importance of visualization. The quiz tests understanding of steps and variable states.