Contents
How do Shapley values work?
The Shapley value is a solution concept used in game theory that involves fairly distributing both gains and costs to several actors working in coalition. The Shapley value ensures each actor gains as much or more as they would have from acting independently.
What is data Shapley?
Given a learning algorithm trained on n data points to produce a predictor, we propose data Shapley as a metric to quantify the value of each training datum to the predictor performance. Data Shapley uniquely satisfies several natural properties of equitable data valuation.
How do you read a shap value?
How to interpret the shap summary plot?
- The y-axis indicates the variable name, in order of importance from top to bottom. The value next to them is the mean SHAP value.
- On the x-axis is the SHAP value.
- Gradient color indicates the original value for that variable.
- Each point represents a row from the original dataset.
What is your data worth equitable data valuation in machine learning?
Given a learning algorithm trained on n data points to produce a predictor, we study data Shapley as an equitable metric to quantify the value of each training datum to the predictor performance.
What is data asset valuation?
Data Valuation is an emerging discipline aimed at estimating the information’s value as a business asset. Information is an intangible asset that is becoming more and more important within organizations. Business information and its physical representation in systems are two sides of the same coin.
Which is a characteristic of the Shapley value?
A crucial characteristic of Shapley values is that players’ contributions always add up to the final payoff: 21.66% + 21.66% + 46.66% = 90%. The relevance of this framework to machine learning is apparent if you translate payoff to prediction and players to features.
How are Shapley values used in machine learning?
In machine learning the participants are the features of your input and the collective payout is the model prediction. In that context Shapley values are used to calculate how much each individual feature contributes to the model output. How are Shapley values calculated? Most of the time you tend to see this equation across literature:
Which is better data Shapley or random valuation?
… points that data Shapley considers valuable are crucially important for the model performance while leave-one-out valuation is only slightly better than random valuation (i.e., removing random points). If you go in the other direction, and start by removing the lowest valued data points, the model performance improves!
What is the Shapley value in game theory?
In game theory this is called the Shapley value, so here the authors call it the data Shapley value. Let’s unpick that equation a little: Sum all of those value increments, and divide by the number of subsets to yield the average value increment for a subset of the training data when data point is added to it.