Contextual compression in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Contextual compression

Which metric matters for Contextual Compression and WHY

Contextual compression reduces data size while keeping important meaning. The key metric is Reconstruction Quality, often measured by Perplexity or BLEU score in language tasks. This shows how well the compressed data can be restored or understood. Another important metric is Compression Ratio, which tells how much smaller the data became. We want a good balance: high quality with strong compression.

Confusion Matrix or Equivalent Visualization

Contextual compression is not a classification task, so no confusion matrix applies. Instead, we use a quality vs compression table or graph. For example:

    +----------------+------------------+
    | Compression %  | Reconstruction   |
    | (smaller is    | Quality (e.g.,   |
    | better)        | BLEU score)      |
    +----------------+------------------+
    | 50%            | 0.85             |
    | 30%            | 0.75             |
    | 20%            | 0.60             |
    +----------------+------------------+
    

This shows how quality drops as compression increases.

Precision vs Recall Tradeoff (or Equivalent)

In contextual compression, the tradeoff is between Compression Ratio and Reconstruction Quality. Compressing more saves space but risks losing important details. Compressing less keeps more meaning but uses more space.

For example, compressing a chat history too much might lose key context, making replies less accurate. Compressing lightly keeps context but costs more storage.

What "Good" vs "Bad" Metric Values Look Like

Good: Compression ratio around 30-50% with reconstruction quality (BLEU or similar) above 0.8 means data is much smaller but still clear.

Bad: Compression ratio below 20% with quality below 0.6 means too much info lost, making the compressed data useless.

Common Metrics Pitfalls

• Ignoring quality: Focusing only on compression ratio can lead to unusable data.
• Overfitting compression: Compressing too well on training data but failing on new data.
• Data leakage: Using future context in compression can give unrealistic quality.
• Misleading metrics: Using accuracy or classification metrics instead of reconstruction quality.

Self Check

Your compression model reduces data size by 70% but the reconstruction quality BLEU score is 0.4. Is it good for production? Why or why not?

Answer: No, it is not good. Although the data is much smaller, the low BLEU score means the compressed data loses too much meaning. This will hurt any task relying on the compressed context.

Key Result

Contextual compression balances compression ratio and reconstruction quality to keep meaning while reducing size.

Practice

1. What is the main goal of contextual compression in AI?

easy

A. Keep only the most important information to save space and time

B. Increase the size of the data for better accuracy

C. Remove all data except the first sentence

D. Add random noise to the data to improve learning

Solution

1. Step 1: Understand the purpose of contextual compression

   Contextual compression aims to reduce data size by keeping only key information.

2. Step 2: Compare options with this purpose

   Only Keep only the most important information to save space and time matches this goal by saving space and time through important info retention.

3. Final Answer:

   Keep only the most important information to save space and time -> Option A

4. Quick Check:

   Contextual compression = Keep important info [OK]

Hint: Remember: compression means keeping key info, not deleting all [OK]

Common Mistakes:

• Thinking compression means deleting everything
• Confusing compression with data expansion
• Assuming random data removal improves results

2. Which of the following is the correct way to describe a simple contextual compression method?

easy

A. Remove all punctuation from the text

B. Select key sentences and remove less useful details

C. Translate text into another language

D. Add extra words to make text longer

Solution

1. Step 1: Identify what simple contextual compression does

   It selects important parts and removes less useful details to reduce size.

2. Step 2: Match options to this description

   Select key sentences and remove less useful details correctly describes selecting key sentences and removing less useful details.

3. Final Answer:

   Select key sentences and remove less useful details -> Option B

4. Quick Check:

   Simple compression = select key parts [OK]

Hint: Focus on keeping key parts, not random removal [OK]

Common Mistakes:

• Confusing compression with translation
• Thinking punctuation removal equals compression
• Adding words instead of removing

3. Given the following text: 'The cat sat on the mat. It was sunny outside. The dog barked loudly.' Which compressed version best shows contextual compression?

medium

A. 'It was sunny outside. The dog barked loudly.'

B. 'The dog barked loudly.'

C. 'The cat sat on the mat. It was sunny outside. The dog barked loudly.'

D. 'The cat sat on the mat. The dog barked loudly.'

Solution

1. Step 1: Identify key information in the text

   The cat sitting and the dog barking are key events; the weather is less important.

2. Step 2: Choose the option that keeps key info and removes less useful details

   'The cat sat on the mat. The dog barked loudly.' keeps the cat and dog events, removing the less important weather sentence.

3. Final Answer:

   'The cat sat on the mat. The dog barked loudly.' -> Option D

4. Quick Check:

   Keep key events, drop less useful info = 'The cat sat on the mat. The dog barked loudly.' [OK]

Hint: Keep main events, drop side details [OK]

Common Mistakes:

• Keeping all sentences without compression
• Removing too much and losing key info
• Choosing only one sentence when more is needed

4. You have a compression function that removes all sentences containing the word 'not'. The input is: 'I do not like rain. The sun is bright. It is not cold.' What is the output?

medium

A. '' (empty string)

B. 'I do not like rain. It is not cold.'

C. 'The sun is bright.'

D. 'I do not like rain. The sun is bright. It is not cold.'

Solution

1. Step 1: Identify sentences containing 'not'

   Sentences 1 and 3 contain 'not' and should be removed.

2. Step 2: Remove those sentences and keep the rest

   Only 'The sun is bright.' remains after removal.

3. Final Answer:

   'The sun is bright.' -> Option C

4. Quick Check:

   Remove 'not' sentences = 'The sun is bright.' [OK]

Hint: Remove sentences with 'not' only [OK]

Common Mistakes:

• Keeping sentences with 'not'
• Removing all sentences
• Returning original text unchanged

5. You want to compress a conversation by keeping only sentences with keywords: ['urgent', 'meeting', 'deadline']. Given the conversation: 'We have a meeting tomorrow. The weather is nice. The deadline is next week. Let's grab lunch.' Which compressed output is correct?

hard

A. 'We have a meeting tomorrow. The deadline is next week.'

B. 'The weather is nice. Let's grab lunch.'

C. 'We have a meeting tomorrow. The weather is nice.'

D. 'Let's grab lunch. The deadline is next week.'

Solution

1. Step 1: Identify sentences containing keywords

   Sentences with 'meeting' and 'deadline' are the first and third sentences.

2. Step 2: Keep only those sentences and remove others

   Keep 'We have a meeting tomorrow.' and 'The deadline is next week.'

3. Final Answer:

   'We have a meeting tomorrow. The deadline is next week.' -> Option A

4. Quick Check:

   Keep keyword sentences = 'We have a meeting tomorrow. The deadline is next week.' [OK]

Hint: Keep sentences with keywords only [OK]

Common Mistakes:

• Keeping sentences without keywords
• Removing all sentences
• Mixing unrelated sentences