Genetic algorithms reveal profound individual differences in emotion recognition -- Nicola Binetti, Nadejda Roubtsova, Christina Carlisi, Darren Cosker, Essi Viding and Isabelle Mareschal CSV Files: -- 1) PreferredExpressions.csv. Generations of expressions evolved with the GA tool by participants online (1-293) and in a controlled laboratory environment (294-336). The csv contains 149 blendshapes corresponding to the preferred expression (labelled as "most compatible with target emotion") flagged within each of 10 generations of the Genetic Algorithm tool. The last columns indicate Subject, Generation number and Emotion. ZIP Files: -- 1) GA Expression - Data. 10 Generations of expressions evolved with the GA tool by participants online (1-293) and in a controlled laboratory environment (294-336). The Preferred expressions corresponds to the expression in the last generation that was labelled as "most compatible with target emotion". Each .csv file contains an expression type (happy, fear, sad or angry) that was evolved by a participant throughout 11 iterations of the GA tool. The filename specifies the participant ID (1-336), the Block number (1-4) and the target expression. Columns / rows blendshapes 1 to 150: the list of blendshapes that control the avatar expression (columns A through ET). Each row provides the blendhsape weights (between 0 and 1) that correspond to each expression. Note that the analyses were run on 149 blendshapes, as the 1st blendshape (Neutral) does not vary. faceID: an identifier of expressions presented on a given trial (on each trial 10 faces are presented on two rows. Each face is labelled with a number 1-10. Participants make their selections by specifying the corresponding numbers of expressions that are compatible with the target category). faceSelected: The faces that were selected within the display as being "compatible with the target expression" (flagged as "1"). Participants can select as few as 1 expression and as many as 10 expressions. eliteSelected: the face amongst the selection of compatible expressions, that was selected as the closest fit of the target expression (flagged as "1"). Participants can only select 1 face. nGeneration: the iteration of the GA algorithm. Note that only from the 2nd iteration (labelled as nGeneration 2) expressions are evolved based on participant selections. 2) GA Expression - Images - expression renders of participants' Preferred expression. Note that order does not coincide with participant number IDs, as stored in GA Expression - Data 3) Dataset Preferred expression Validation - two datasets: 1) with interleaved GA/KDEF stimuli (randomly presented on each trial), 2) with blocked GA/KDEF (presented in separate blocks)). 4) Dataset Preferred expression categorization task. Note that participants also evolved GA expressions (their evolved expression data is found in GA Expression - Data batches 3 & 4). 5) Dataset Preferred expression discrimination task. Note that participants also evolved GA expressions (their evolved expression data is found in GA Expression - Data). Each file specifies pair number (31 pairs) and participant number in GA Expression - Data. 6) BlendshapeAnalysis.ipynb . Evolution of selected expression blendshapes across GA generations. 7) PreferredExpressionAnalysis.ipynb . mapping of expressions and clustering of Preferred expressions 8) InterSubjectEliteDifferences.ipynb . Distribution of differences (cosine distance in blendshape space) of all possible Preferred expression pairings per emotion category. 9) PredictEmotionSVM.ipynb . Classify emotion category of Preferred facial expressions via Support Vector Machines SVM 10) RankActivations.ipynb . Rank Preferred expression blendshapes based on peak activation or peak variability. 11) PreferredExpressionGMM.py . Preferred expressions per emotion cluster, identified through Gaussian Mixture Model 12) ExpressionCategorizationAnalysis.py . Analysis of expression categorization data 13) ExpressionDiscriminationAnalysis.ipynb . Analysis of expression discrimination data code: -- 1. The GA tool is included in code.zip 2. Code Documentation is included in code_documentation.zip 3. We also offer a Github repo at: https://github.com/alexs7/Emogen-Public-OSF 4. We will release an on-line version of the GA tool that can be run from within a browser window. The GA tool client website will become live around January 2023. We will provide further information at a GA dissemination meeting that will be held in London (Queen Mary University of London) in December 2022. cite: -- If you use our toolkit, please cite our paper: Binetti, N., Roubtsova, N., Carlisi, C., Cosker, D., Viding, E. & Mareschal I. (2022). Genetic algorithms reveal profound individual differences in emotion recognition. Proceedings of the National Academy of Science, 119 (45), e2201380119 contact: -- nicolabinetti@gmail.com i.mareschal@qmul.ac.uk ar2056@bath.ac.uk (for questions concerning the GA tool installation / running) coming soon: -- January 2023, an online version of the GA tool