PeerJ Computer Science:Data Mining and Machine Learninghttps://peerj.com/articles/index.atom?journal=cs&subject=9500Data Mining and Machine Learning articles published in PeerJ Computer ScienceAdaptations of data mining methodologies: a systematic literature reviewhttps://peerj.com/articles/002672020-05-252020-05-25Veronika PlotnikovaMarlon DumasFredrik Milani
The use of end-to-end data mining methodologies such as CRISP-DM, KDD process, and SEMMA has grown substantially over the past decade. However, little is known as to how these methodologies are used in practice. In particular, the question of whether data mining methodologies are used ‘as-is’ or adapted for specific purposes, has not been thoroughly investigated. This article addresses this gap via a systematic literature review focused on the context in which data mining methodologies are used and the adaptations they undergo. The literature review covers 207 peer-reviewed and ‘grey’ publications. We find that data mining methodologies are primarily applied ‘as-is’. At the same time, we also identify various adaptations of data mining methodologies and we note that their number is growing rapidly. The dominant adaptations pattern is related to methodology adjustments at a granular level (modifications) followed by extensions of existing methodologies with additional elements. Further, we identify two recurrent purposes for adaptation: (1) adaptations to handle Big Data technologies, tools and environments (technological adaptations); and (2) adaptations for context-awareness and for integrating data mining solutions into business processes and IT systems (organizational adaptations). The study suggests that standard data mining methodologies do not pay sufficient attention to deployment issues, which play a prominent role when turning data mining models into software products that are integrated into the IT architectures and business processes of organizations. We conclude that refinements of existing methodologies aimed at combining data, technological, and organizational aspects, could help to mitigate these gaps.
The use of end-to-end data mining methodologies such as CRISP-DM, KDD process, and SEMMA has grown substantially over the past decade. However, little is known as to how these methodologies are used in practice. In particular, the question of whether data mining methodologies are used ‘as-is’ or adapted for specific purposes, has not been thoroughly investigated. This article addresses this gap via a systematic literature review focused on the context in which data mining methodologies are used and the adaptations they undergo. The literature review covers 207 peer-reviewed and ‘grey’ publications. We find that data mining methodologies are primarily applied ‘as-is’. At the same time, we also identify various adaptations of data mining methodologies and we note that their number is growing rapidly. The dominant adaptations pattern is related to methodology adjustments at a granular level (modifications) followed by extensions of existing methodologies with additional elements. Further, we identify two recurrent purposes for adaptation: (1) adaptations to handle Big Data technologies, tools and environments (technological adaptations); and (2) adaptations for context-awareness and for integrating data mining solutions into business processes and IT systems (organizational adaptations). The study suggests that standard data mining methodologies do not pay sufficient attention to deployment issues, which play a prominent role when turning data mining models into software products that are integrated into the IT architectures and business processes of organizations. We conclude that refinements of existing methodologies aimed at combining data, technological, and organizational aspects, could help to mitigate these gaps.Influence of tweets and diversification on serendipitous research paper recommender systemshttps://peerj.com/articles/cs-2732020-05-182020-05-18Chifumi NishiokaJörn HaukeAnsgar Scherp
In recent years, a large body of literature has accumulated around the topic of research paper recommender systems. However, since most studies have focused on the variable of accuracy, they have overlooked the serendipity of recommendations, which is an important determinant of user satisfaction. Serendipity is concerned with the relevance and unexpectedness of recommendations, and so serendipitous items are considered those which positively surprise users. The purpose of this article was to examine two key research questions: firstly, whether a user’s Tweets can assist in generating more serendipitous recommendations; and secondly, whether the diversification of a list of recommended items further improves serendipity. To investigate these issues, an online experiment was conducted in the domain of computer science with 22 subjects. As an evaluation metric, we use the serendipity score (SRDP), in which the unexpectedness of recommendations is inferred by using a primitive recommendation strategy. The results indicate that a user’s Tweets do not improve serendipity, but they can reflect recent research interests and are typically heterogeneous. Contrastingly, diversification was found to lead to a greater number of serendipitous research paper recommendations.
In recent years, a large body of literature has accumulated around the topic of research paper recommender systems. However, since most studies have focused on the variable of accuracy, they have overlooked the serendipity of recommendations, which is an important determinant of user satisfaction. Serendipity is concerned with the relevance and unexpectedness of recommendations, and so serendipitous items are considered those which positively surprise users. The purpose of this article was to examine two key research questions: firstly, whether a user’s Tweets can assist in generating more serendipitous recommendations; and secondly, whether the diversification of a list of recommended items further improves serendipity. To investigate these issues, an online experiment was conducted in the domain of computer science with 22 subjects. As an evaluation metric, we use the serendipity score (SRDP), in which the unexpectedness of recommendations is inferred by using a primitive recommendation strategy. The results indicate that a user’s Tweets do not improve serendipity, but they can reflect recent research interests and are typically heterogeneous. Contrastingly, diversification was found to lead to a greater number of serendipitous research paper recommendations.Adaptive divergence for rapid adversarial optimizationhttps://peerj.com/articles/cs-2742020-05-182020-05-18Maxim BorisyakTatiana GaintsevaAndrey Ustyuzhanin
Adversarial Optimization provides a reliable, practical way to match two implicitly defined distributions, one of which is typically represented by a sample of real data, and the other is represented by a parameterized generator. Matching of the distributions is achieved by minimizing a divergence between these distribution, and estimation of the divergence involves a secondary optimization task, which, typically, requires training a model to discriminate between these distributions. The choice of the model has its trade-off: high-capacity models provide good estimations of the divergence, but, generally, require large sample sizes to be properly trained. In contrast, low-capacity models tend to require fewer samples for training; however, they might provide biased estimations. Computational costs of Adversarial Optimization becomes significant when sampling from the generator is expensive. One of the practical examples of such settings is fine-tuning parameters of complex computer simulations. In this work, we introduce a novel family of divergences that enables faster optimization convergence measured by the number of samples drawn from the generator. The variation of the underlying discriminator model capacity during optimization leads to a significant speed-up. The proposed divergence family suggests using low-capacity models to compare distant distributions (typically, at early optimization steps), and the capacity gradually grows as the distributions become closer to each other. Thus, it allows for a significant acceleration of the initial stages of optimization. This acceleration was demonstrated on two fine-tuning problems involving Pythia event generator and two of the most popular black-box optimization algorithms: Bayesian Optimization and Variational Optimization. Experiments show that, given the same budget, adaptive divergences yield results up to an order of magnitude closer to the optimum than Jensen-Shannon divergence. While we consider physics-related simulations, adaptive divergences can be applied to any stochastic simulation.
Adversarial Optimization provides a reliable, practical way to match two implicitly defined distributions, one of which is typically represented by a sample of real data, and the other is represented by a parameterized generator. Matching of the distributions is achieved by minimizing a divergence between these distribution, and estimation of the divergence involves a secondary optimization task, which, typically, requires training a model to discriminate between these distributions. The choice of the model has its trade-off: high-capacity models provide good estimations of the divergence, but, generally, require large sample sizes to be properly trained. In contrast, low-capacity models tend to require fewer samples for training; however, they might provide biased estimations. Computational costs of Adversarial Optimization becomes significant when sampling from the generator is expensive. One of the practical examples of such settings is fine-tuning parameters of complex computer simulations. In this work, we introduce a novel family of divergences that enables faster optimization convergence measured by the number of samples drawn from the generator. The variation of the underlying discriminator model capacity during optimization leads to a significant speed-up. The proposed divergence family suggests using low-capacity models to compare distant distributions (typically, at early optimization steps), and the capacity gradually grows as the distributions become closer to each other. Thus, it allows for a significant acceleration of the initial stages of optimization. This acceleration was demonstrated on two fine-tuning problems involving Pythia event generator and two of the most popular black-box optimization algorithms: Bayesian Optimization and Variational Optimization. Experiments show that, given the same budget, adaptive divergences yield results up to an order of magnitude closer to the optimum than Jensen-Shannon divergence. While we consider physics-related simulations, adaptive divergences can be applied to any stochastic simulation.SANgo: a storage infrastructure simulator with reinforcement learning supporthttps://peerj.com/articles/cs-2712020-05-042020-05-04Kenenbek ArzymatovAndrey SapronovVladislav BelavinLeonid GremyachikhMaksim KarpovAndrey UstyuzhaninIvan TchoubArtem Ikoev
We introduce SANgo (Storage Area Network in the Go language)—a Go-based package for simulating the behavior of modern storage infrastructure. The software is based on the discrete-event modeling paradigm and captures the structure and dynamics of high-level storage system building blocks. The flexible structure of the package allows us to create a model of a real storage system with a configurable number of components. The granularity of the simulated system can be defined depending on the replicated patterns of actual system behavior. Accurate replication enables us to reach the primary goal of our simulator—to explore the stability boundaries of real storage systems. To meet this goal, SANgo offers a variety of interfaces for easy monitoring and tuning of the simulated model. These interfaces allow us to track the number of metrics of such components as storage controllers, network connections, and hard-drives. Other interfaces allow altering the parameter values of the simulated system effectively in real-time, thus providing the possibility for training a realistic digital twin using, for example, the reinforcement learning (RL) approach. One can train an RL model to reduce discrepancies between simulated and real SAN data. The external control algorithm can adjust the simulator parameters to make the difference as small as possible. SANgo supports the standard OpenAI gym interface; thus, the software can serve as a benchmark for comparison of different learning algorithms.
We introduce SANgo (Storage Area Network in the Go language)—a Go-based package for simulating the behavior of modern storage infrastructure. The software is based on the discrete-event modeling paradigm and captures the structure and dynamics of high-level storage system building blocks. The flexible structure of the package allows us to create a model of a real storage system with a configurable number of components. The granularity of the simulated system can be defined depending on the replicated patterns of actual system behavior. Accurate replication enables us to reach the primary goal of our simulator—to explore the stability boundaries of real storage systems. To meet this goal, SANgo offers a variety of interfaces for easy monitoring and tuning of the simulated model. These interfaces allow us to track the number of metrics of such components as storage controllers, network connections, and hard-drives. Other interfaces allow altering the parameter values of the simulated system effectively in real-time, thus providing the possibility for training a realistic digital twin using, for example, the reinforcement learning (RL) approach. One can train an RL model to reduce discrepancies between simulated and real SAN data. The external control algorithm can adjust the simulator parameters to make the difference as small as possible. SANgo supports the standard OpenAI gym interface; thus, the software can serve as a benchmark for comparison of different learning algorithms.A comparative study of machine learning and deep learning algorithms to classify cancer types based on microarray gene expression datahttps://peerj.com/articles/cs-2702020-04-132020-04-13Reinel Tabares-SotoSimon Orozco-AriasVictor Romero-CanoVanesa Segovia BucheliJosé Luis Rodríguez-SoteloCristian Felipe Jiménez-Varón
Cancer classification is a topic of major interest in medicine since it allows accurate and efficient diagnosis and facilitates a successful outcome in medical treatments. Previous studies have classified human tumors using a large-scale RNA profiling and supervised Machine Learning (ML) algorithms to construct a molecular-based classification of carcinoma cells from breast, bladder, adenocarcinoma, colorectal, gastro esophagus, kidney, liver, lung, ovarian, pancreas, and prostate tumors. These datasets are collectively known as the 11_tumor database, although this database has been used in several works in the ML field, no comparative studies of different algorithms can be found in the literature. On the other hand, advances in both hardware and software technologies have fostered considerable improvements in the precision of solutions that use ML, such as Deep Learning (DL). In this study, we compare the most widely used algorithms in classical ML and DL to classify the tumors described in the 11_tumor database. We obtained tumor identification accuracies between 90.6% (Logistic Regression) and 94.43% (Convolutional Neural Networks) using k-fold cross-validation. Also, we show how a tuning process may or may not significantly improve algorithms’ accuracies. Our results demonstrate an efficient and accurate classification method based on gene expression (microarray data) and ML/DL algorithms, which facilitates tumor type prediction in a multi-cancer-type scenario.
Cancer classification is a topic of major interest in medicine since it allows accurate and efficient diagnosis and facilitates a successful outcome in medical treatments. Previous studies have classified human tumors using a large-scale RNA profiling and supervised Machine Learning (ML) algorithms to construct a molecular-based classification of carcinoma cells from breast, bladder, adenocarcinoma, colorectal, gastro esophagus, kidney, liver, lung, ovarian, pancreas, and prostate tumors. These datasets are collectively known as the 11_tumor database, although this database has been used in several works in the ML field, no comparative studies of different algorithms can be found in the literature. On the other hand, advances in both hardware and software technologies have fostered considerable improvements in the precision of solutions that use ML, such as Deep Learning (DL). In this study, we compare the most widely used algorithms in classical ML and DL to classify the tumors described in the 11_tumor database. We obtained tumor identification accuracies between 90.6% (Logistic Regression) and 94.43% (Convolutional Neural Networks) using k-fold cross-validation. Also, we show how a tuning process may or may not significantly improve algorithms’ accuracies. Our results demonstrate an efficient and accurate classification method based on gene expression (microarray data) and ML/DL algorithms, which facilitates tumor type prediction in a multi-cancer-type scenario.A survey on exponential random graph models: an application perspectivehttps://peerj.com/articles/cs-2692020-04-062020-04-06Saeid GhafouriSeyed Hossein Khasteh
The uncertainty underlying real-world phenomena has attracted attention toward statistical analysis approaches. In this regard, many problems can be modeled as networks. Thus, the statistical analysis of networked problems has received special attention from many researchers in recent years. Exponential Random Graph Models, known as ERGMs, are one of the popular statistical methods for analyzing the graphs of networked data. ERGM is a generative statistical network model whose ultimate goal is to present a subset of networks with particular characteristics as a statistical distribution. In the context of ERGMs, these graph’s characteristics are called statistics or configurations. Most of the time they are the number of repeated subgraphs across the graphs. Some examples include the number of triangles or the number of cycle of an arbitrary length. Also, any other census of the graph, as with the edge density, can be considered as one of the graph’s statistics. In this review paper, after explaining the building blocks and classic methods of ERGMs, we have reviewed their newly presented approaches and research papers. Further, we have conducted a comprehensive study on the applications of ERGMs in many research areas which to the best of our knowledge has not been done before. This review paper can be used as an introduction for scientists from various disciplines whose aim is to use ERGMs in some networked data in their field of expertise.
The uncertainty underlying real-world phenomena has attracted attention toward statistical analysis approaches. In this regard, many problems can be modeled as networks. Thus, the statistical analysis of networked problems has received special attention from many researchers in recent years. Exponential Random Graph Models, known as ERGMs, are one of the popular statistical methods for analyzing the graphs of networked data. ERGM is a generative statistical network model whose ultimate goal is to present a subset of networks with particular characteristics as a statistical distribution. In the context of ERGMs, these graph’s characteristics are called statistics or configurations. Most of the time they are the number of repeated subgraphs across the graphs. Some examples include the number of triangles or the number of cycle of an arbitrary length. Also, any other census of the graph, as with the edge density, can be considered as one of the graph’s statistics. In this review paper, after explaining the building blocks and classic methods of ERGMs, we have reviewed their newly presented approaches and research papers. Further, we have conducted a comprehensive study on the applications of ERGMs in many research areas which to the best of our knowledge has not been done before. This review paper can be used as an introduction for scientists from various disciplines whose aim is to use ERGMs in some networked data in their field of expertise.An evolutionary decomposition-based multi-objective feature selection for multi-label classificationhttps://peerj.com/articles/cs-2612020-03-022020-03-02Azam Asilian BidgoliHossein Ebrahimpour-KomlehShahryar Rahnamayan
Data classification is a fundamental task in data mining. Within this field, the classification of multi-labeled data has been seriously considered in recent years. In such problems, each data entity can simultaneously belong to several categories. Multi-label classification is important because of many recent real-world applications in which each entity has more than one label. To improve the performance of multi-label classification, feature selection plays an important role. It involves identifying and removing irrelevant and redundant features that unnecessarily increase the dimensions of the search space for the classification problems. However, classification may fail with an extreme decrease in the number of relevant features. Thus, minimizing the number of features and maximizing the classification accuracy are two desirable but conflicting objectives in multi-label feature selection. In this article, we introduce a multi-objective optimization algorithm customized for selecting the features of multi-label data. The proposed algorithm is an enhanced variant of a decomposition-based multi-objective optimization approach, in which the multi-label feature selection problem is divided into single-objective subproblems that can be simultaneously solved using an evolutionary algorithm. This approach leads to accelerating the optimization process and finding more diverse feature subsets. The proposed method benefits from a local search operator to find better solutions for each subproblem. We also define a pool of genetic operators to generate new feature subsets based on old generation. To evaluate the performance of the proposed algorithm, we compare it with two other multi-objective feature selection approaches on eight real-world benchmark datasets that are commonly used for multi-label classification. The reported results of multi-objective method evaluation measures, such as hypervolume indicator and set coverage, illustrate an improvement in the results obtained by the proposed method. Moreover, the proposed method achieved better results in terms of classification accuracy with fewer features compared with state-of-the-art methods.
Data classification is a fundamental task in data mining. Within this field, the classification of multi-labeled data has been seriously considered in recent years. In such problems, each data entity can simultaneously belong to several categories. Multi-label classification is important because of many recent real-world applications in which each entity has more than one label. To improve the performance of multi-label classification, feature selection plays an important role. It involves identifying and removing irrelevant and redundant features that unnecessarily increase the dimensions of the search space for the classification problems. However, classification may fail with an extreme decrease in the number of relevant features. Thus, minimizing the number of features and maximizing the classification accuracy are two desirable but conflicting objectives in multi-label feature selection. In this article, we introduce a multi-objective optimization algorithm customized for selecting the features of multi-label data. The proposed algorithm is an enhanced variant of a decomposition-based multi-objective optimization approach, in which the multi-label feature selection problem is divided into single-objective subproblems that can be simultaneously solved using an evolutionary algorithm. This approach leads to accelerating the optimization process and finding more diverse feature subsets. The proposed method benefits from a local search operator to find better solutions for each subproblem. We also define a pool of genetic operators to generate new feature subsets based on old generation. To evaluate the performance of the proposed algorithm, we compare it with two other multi-objective feature selection approaches on eight real-world benchmark datasets that are commonly used for multi-label classification. The reported results of multi-objective method evaluation measures, such as hypervolume indicator and set coverage, illustrate an improvement in the results obtained by the proposed method. Moreover, the proposed method achieved better results in terms of classification accuracy with fewer features compared with state-of-the-art methods.Image-based effective feature generation for protein structural class and ligand binding predictionhttps://peerj.com/articles/cs-2532020-02-032020-02-03Nafees SadiqueAl Amin Neaz AhmedMd Tajul IslamMd. Nawshad PervageSwakkhar Shatabda
Proteins are the building blocks of all cells in both human and all living creatures of the world. Most of the work in the living organism is performed by proteins. Proteins are polymers of amino acid monomers which are biomolecules or macromolecules. The tertiary structure of protein represents the three-dimensional shape of a protein. The functions, classification and binding sites are governed by the protein’s tertiary structure. If two protein structures are alike, then the two proteins can be of the same kind implying similar structural class and ligand binding properties. In this paper, we have used the protein tertiary structure to generate effective features for applications in structural similarity to detect structural class and ligand binding. Firstly, we have analyzed the effectiveness of a group of image-based features to predict the structural class of a protein. These features are derived from the image generated by the distance matrix of the tertiary structure of a given protein. They include local binary pattern (LBP) histogram, Gabor filtered LBP histogram, separate row multiplication matrix with uniform LBP histogram, neighbor block subtraction matrix with uniform LBP histogram and atom bond. Separate row multiplication matrix and neighbor block subtraction matrix filters, as well as atom bond, are our novels. The experiments were done on a standard benchmark dataset. We have demonstrated the effectiveness of these features over a large variety of supervised machine learning algorithms. Experiments suggest support vector machines is the best performing classifier on the selected dataset using the set of features. We believe the excellent performance of Hybrid LBP in terms of accuracy would motivate the researchers and practitioners to use it to identify protein structural class. To facilitate that, a classification model using Hybrid LBP is readily available for use at http://brl.uiu.ac.bd/PL/. Protein-ligand binding is accountable for managing the tasks of biological receptors that help to cure diseases and many more. Therefore, binding prediction between protein and ligand is important for understanding a protein’s activity or to accelerate docking computations in virtual screening-based drug design. Protein-ligand binding prediction requires three-dimensional tertiary structure of the target protein to be searched for ligand binding. In this paper, we have proposed a supervised learning algorithm for predicting protein-ligand binding, which is a similarity-based clustering approach using the same set of features. Our algorithm works better than the most popular and widely used machine learning algorithms.
Proteins are the building blocks of all cells in both human and all living creatures of the world. Most of the work in the living organism is performed by proteins. Proteins are polymers of amino acid monomers which are biomolecules or macromolecules. The tertiary structure of protein represents the three-dimensional shape of a protein. The functions, classification and binding sites are governed by the protein’s tertiary structure. If two protein structures are alike, then the two proteins can be of the same kind implying similar structural class and ligand binding properties. In this paper, we have used the protein tertiary structure to generate effective features for applications in structural similarity to detect structural class and ligand binding. Firstly, we have analyzed the effectiveness of a group of image-based features to predict the structural class of a protein. These features are derived from the image generated by the distance matrix of the tertiary structure of a given protein. They include local binary pattern (LBP) histogram, Gabor filtered LBP histogram, separate row multiplication matrix with uniform LBP histogram, neighbor block subtraction matrix with uniform LBP histogram and atom bond. Separate row multiplication matrix and neighbor block subtraction matrix filters, as well as atom bond, are our novels. The experiments were done on a standard benchmark dataset. We have demonstrated the effectiveness of these features over a large variety of supervised machine learning algorithms. Experiments suggest support vector machines is the best performing classifier on the selected dataset using the set of features. We believe the excellent performance of Hybrid LBP in terms of accuracy would motivate the researchers and practitioners to use it to identify protein structural class. To facilitate that, a classification model using Hybrid LBP is readily available for use at http://brl.uiu.ac.bd/PL/. Protein-ligand binding is accountable for managing the tasks of biological receptors that help to cure diseases and many more. Therefore, binding prediction between protein and ligand is important for understanding a protein’s activity or to accelerate docking computations in virtual screening-based drug design. Protein-ligand binding prediction requires three-dimensional tertiary structure of the target protein to be searched for ligand binding. In this paper, we have proposed a supervised learning algorithm for predicting protein-ligand binding, which is a similarity-based clustering approach using the same set of features. Our algorithm works better than the most popular and widely used machine learning algorithms.Beyond the topics: how deep learning can improve the discriminability of probabilistic topic modellinghttps://peerj.com/articles/cs-2522020-01-272020-01-27Noura Al MoubayedStephen McGoughBashar Awwad Shiekh Hasan
The article presents a discriminative approach to complement the unsupervised probabilistic nature of topic modelling. The framework transforms the probabilities of the topics per document into class-dependent deep learning models that extract highly discriminatory features suitable for classification. The framework is then used for sentiment analysis with minimum feature engineering. The approach transforms the sentiment analysis problem from the word/document domain to the topics domain making it more robust to noise and incorporating complex contextual information that are not represented otherwise. A stacked denoising autoencoder (SDA) is then used to model the complex relationship among the topics per sentiment with minimum assumptions. To achieve this, a distinct topic model and SDA per sentiment polarity is built with an additional decision layer for classification. The framework is tested on a comprehensive collection of benchmark datasets that vary in sample size, class bias and classification task. A significant improvement to the state of the art is achieved without the need for a sentiment lexica or over-engineered features. A further analysis is carried out to explain the observed improvement in accuracy.
The article presents a discriminative approach to complement the unsupervised probabilistic nature of topic modelling. The framework transforms the probabilities of the topics per document into class-dependent deep learning models that extract highly discriminatory features suitable for classification. The framework is then used for sentiment analysis with minimum feature engineering. The approach transforms the sentiment analysis problem from the word/document domain to the topics domain making it more robust to noise and incorporating complex contextual information that are not represented otherwise. A stacked denoising autoencoder (SDA) is then used to model the complex relationship among the topics per sentiment with minimum assumptions. To achieve this, a distinct topic model and SDA per sentiment polarity is built with an additional decision layer for classification. The framework is tested on a comprehensive collection of benchmark datasets that vary in sample size, class bias and classification task. A significant improvement to the state of the art is achieved without the need for a sentiment lexica or over-engineered features. A further analysis is carried out to explain the observed improvement in accuracy.Machine learning of symbolic compositional rules with genetic programming: dissonance treatment in Palestrinahttps://peerj.com/articles/cs-2442019-12-162019-12-16Torsten AndersBenjamin Inden
We describe a method for automatically extracting symbolic compositional rules from music corpora. Resulting rules are expressed by a combination of logic and numeric relations, and they can therefore be studied by humans. These rules can also be used for algorithmic composition, where they can be combined with each other and with manually programmed rules. We chose genetic programming (GP) as our machine learning technique, because it is capable of learning formulas consisting of both logic and numeric relations. GP was never used for this purpose to our knowledge. We therefore investigate a well understood case in this study: dissonance treatment in Palestrina’s music. We label dissonances with a custom algorithm, automatically cluster melodic fragments with labelled dissonances into different dissonance categories (passing tone, suspension etc.) with the DBSCAN algorithm, and then learn rules describing the dissonance treatment of each category with GP. Learning is based on the requirement that rules must be broad enough to cover positive examples, but narrow enough to exclude negative examples. Dissonances from a given category are used as positive examples, while dissonances from other categories, melodic fragments without dissonances, purely random melodic fragments, and slight random transformations of positive examples, are used as negative examples.
We describe a method for automatically extracting symbolic compositional rules from music corpora. Resulting rules are expressed by a combination of logic and numeric relations, and they can therefore be studied by humans. These rules can also be used for algorithmic composition, where they can be combined with each other and with manually programmed rules. We chose genetic programming (GP) as our machine learning technique, because it is capable of learning formulas consisting of both logic and numeric relations. GP was never used for this purpose to our knowledge. We therefore investigate a well understood case in this study: dissonance treatment in Palestrina’s music. We label dissonances with a custom algorithm, automatically cluster melodic fragments with labelled dissonances into different dissonance categories (passing tone, suspension etc.) with the DBSCAN algorithm, and then learn rules describing the dissonance treatment of each category with GP. Learning is based on the requirement that rules must be broad enough to cover positive examples, but narrow enough to exclude negative examples. Dissonances from a given category are used as positive examples, while dissonances from other categories, melodic fragments without dissonances, purely random melodic fragments, and slight random transformations of positive examples, are used as negative examples.