Ensemble

Dataset: potions

Models

• anthropic
• openai35
• random

Model Narratives

anthropic

Round ID: 280
Prompt used:
	Classification Rules for Entity Effectiveness:
	
	1. High Effectiveness Criteria:
	   - FizzIntensity must be > 45
	   - ColourShift must be > 10
	   
	2. Low Effectiveness Criteria:
	   - FizzIntensity must be < 40
	   - ColourShift must be < 10
	
	3. Borderline Case Scoring:
	   - Calculate a weighted score: 
	     (FizzIntensity * 0.6) + (ColourShift * 0.4)
	   - If score > 35, classify as Effective
	   - If score < 30, classify as Ineffective
	
	4. Intermediate Cases:
	   - If criteria are not clearly met, use the weighted score to determine classification
	   - Carefully evaluate cases with FizzIntensity between 40-45 and ColourShift between 10-15
	
	Classification Decision Process:
	- First check high and low effectiveness criteria
	- If inconclusive, apply weighted scoring
	- Aim to minimize false positives and false negatives by using a nuanced approach

Confusion Matrix:
                Predicted Effective  Predicted Ineffective
Actual Effective                    4                    5
Actual Ineffective                    1                   10

Accuracy: 0.700
Precision: 0.800
Recall: 0.444
F1 Score: 0.571

Examples for Correctly predicted Effective: (Correct answer: Effective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 44.953373
	ColourShift: 12.993897


Examples for Falsely predicted Ineffective when it should have been Effective: (Correct answer: Effective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 36.28945
	ColourShift: 11.461653


Examples for Falsely predicted Effective when it should have been Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 49.864723
	ColourShift: 16.120462


Examples for Correctly predicted Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 41.101128
	ColourShift: 15.34901

openai35

Round ID: 486
Prompt used:
	Prompt: If FizzIntensity is greater than 40, predict as Effective. If FizzIntensity is 40 or lower, predict as Ineffective.

Confusion Matrix:
                Predicted Effective  Predicted Ineffective
Actual Effective                    7                    2
Actual Ineffective                    3                    8

Accuracy: 0.750
Precision: 0.700
Recall: 0.778
F1 Score: 0.737

Examples for Correctly predicted Effective: (Correct answer: Effective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 45.78967
	ColourShift: 12.371225


Examples for Falsely predicted Ineffective when it should have been Effective: (Correct answer: Effective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 36.28945
	ColourShift: 11.461653


Examples for Falsely predicted Effective when it should have been Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 41.101128
	ColourShift: 15.34901


Examples for Correctly predicted Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 26.284357
	ColourShift: 21.189081

random

Round ID: 134
Prompt used:
	Choose randomly

Confusion Matrix:
                Predicted Effective  Predicted Ineffective
Actual Effective                    3                    6
Actual Ineffective                    6                    5

Accuracy: 0.400
Precision: 0.333
Recall: 0.333
F1 Score: 0.333

Examples for Correctly predicted Effective: (Correct answer: Effective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 45.78967
	ColourShift: 12.371225


Examples for Falsely predicted Ineffective when it should have been Effective: (Correct answer: Effective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 37.055344
	ColourShift: 15.48838


Examples for Falsely predicted Effective when it should have been Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Effective)
  Entity Data:
	FizzIntensity: 37.77209
	ColourShift: 14.7023735


Examples for Correctly predicted Ineffective: (Correct answer: Ineffective, What the previous set of rules predicted: Ineffective)
  Entity Data:
	FizzIntensity: 49.864723
	ColourShift: 16.120462

Ensemble Confusion Matrix

Accuracy 0.650, Precision 0.667, Recall 0.444, F1 0.533