{"id":3693,"date":"2019-10-18T10:02:53","date_gmt":"2019-10-18T09:02:53","guid":{"rendered":"https:\/\/www.alvantia.com\/?p=3693"},"modified":"2019-10-18T10:02:57","modified_gmt":"2019-10-18T09:02:57","slug":"meta-algorithms-and-base-models-in-machine-learning","status":"publish","type":"post","link":"https:\/\/www.alvantia.com\/en\/meta-algorithms-and-base-models-in-machine-learning\/","title":{"rendered":"Meta algorithms and base models in Machine Learning"},"content":{"rendered":"\n

A traditional, basic approach in the problems of machine learning<\/span><\/strong> is to use a single determined algorithm or model in the implementation of a valid solution. This traditional method is based on creating a pipeline of all the processes that are covered by the development of this algorithm.<\/p>\n\n\n\n\n\n\n\n

This pipeline consists of several phases:\nETL processing, data analysis, model implementation and training, results\nanalysis based on determined metrics and other purging mechanisms, and lastly,\nthe launching into production of the model.<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

Primarily, during the research and implementation\nphase of the model, one or several determined algorithms are used as a\npotential solution to the problem given its category. Naturally, regression\nalgorithms will be chosen to predict the outcome of a quantitative variable. In\na more elaborate approach, where the number of features is so large that\nmonitored analysis and interpretation of the results takes on a higher degree\nof complexity, it is possible to split the problem into various phases by using\nmeta algorithms.<\/p>\n\n\n\n

In this situation, the cycle of testing,\nadjusting and error correction is fundamental in the search for the algorithm\nor model with the correct hyper-parameters.<\/p>\n\n\n\n

A parallel solution is the composition of base algorithms and a superior meta algorithm.<\/span><\/strong> This set of algorithms forms a stack, where the meta algorithm is the last link in the chain, whereas the base algorithms will be the ones implemented simultaneously to feed their input.<\/p>\n\n\n\n

In this architectural approach, base\nalgorithms use different sets of input features; some of them may be shared. In\nthis way, we will have different models for each set of determined features.\nThe way to separate these sets is based on the prior analysis of the data, or\nfailing this, on the intake source .<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

In practice, a non-trivial problem such as\nthe detection of fraudulent transactions or financial fraud, the amount of\nattributes to be handled can become uncontrollable. Using the meta algorithm\nstructure, a type of algorithm is determined for each data source or set of\ndetermined features and to be able to establish which features are essential or\ncan boost the development of the final model. <\/p>\n\n\n\n

Moreover, different models can be studied\nfor different sets of attributes and be analysed to improve the correlation\nbetween them. This maximises the capacity to select the correct model and set\nof predominant features in the development of the model.<\/p>\n\n\n\n

In this architecture, the meta algorithm\nuses the percentage of precision or metric associated to each of the base\nalgorithms, using the binary label as a target, in this case, true or false.<\/p>\n\n\n\n

In the analysis of meta algorithms it is necessary to use SHAP<\/a><\/strong> (SHapely Additive exPlanations) to determine which of the base algorithms is most relevant to produce actual predictions and labels.<\/p>\n\n\n\n

SHAP is an approach for explaining the\nresults of the models or algorithms of machine learning. It assimilates game\ntheory with local explanation methods to represent possible expectation-based\nfeatures. It detects the existing correlation and the value provided by the features\nin the results of the model.<\/p>\n\n\n\n

In the following example, it is possible to\nobserve which attributes contribute best to the learning model. <\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

The sex\nand age features are implicitly correlated in the model\u2019s inference depending\non their age range. In this example, a man with an age of approximately 60 will\nobtain a greater impact on the model\u2019s development. Whereas a female of the\nsame age will not stand out.<\/p>\n\n\n\n

Going\nback to the case of financial fraud detection, this architecture, apart from\nhaving the advantages shown, has some disadvantages.<\/p>\n\n\n\n

The main\ndisadvantage is presented during the implementation, training and testing\nphase. This is due to the stack of base algorithms to be implemented and\ntaught. Code development becomes fairly complex, training computationally\ncomplicated and we are conditioned by the source and volume of data.<\/p>\n\n\n\n

The time\nnecessary for the development of the entire set of algorithms is much\nlonger-lasting than the implementation of traditional measures.<\/p>\n\n\n\n

However,\nas a phase for research, executing batch processes for processing meta\nalgorithms with different configurations of models and duly programmed sets of\nfeatures, it is possible to achieve solutions to more complex problems.<\/p>\n\n\n\n

As assistance for the development of this architecture, Python\u2019s MLxtend library<\/strong><\/a> has inherited all the Scikit-Learn algorithms<\/strong><\/a> to implement the model stack and feed a principal meta algorithm.<\/p>\n\n\n\n

Due to\nthe complexity of this paradigm, it is necessary to start the research phase\nwith simpler solutions. In the principle of parsimony, more well-known as\nOccam’s razor:<\/p>\n\n\n\n

\u201cAll things being equal, the simplest solution tends to be the best one.\u201c<\/span><\/em><\/strong><\/h5>\n\n\n\n

This\nformula applied in the field of statistical and computational learning theory\ncould be expressed as follows:<\/p>\n\n\n\n

\u201cThe less complex a machine learning model is, the less probable it will be that a good empirical result is simply due to the peculiarities of the sample taken.\u201d<\/span><\/em><\/strong><\/h5>\n
<\/div>","protected":false},"excerpt":{"rendered":"

A traditional, basic approach in the problems of machine learning is to use a single determined algorithm or model in the implementation of a valid solution. This traditional method is based on creating a pipeline of all the processes that are covered by the development of this algorithm.<\/p>\n

Continue Reading<\/i><\/a><\/p>\n","protected":false},"author":5,"featured_media":3603,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[215,186,187],"tags":[37,315,171],"class_list":["post-3693","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-alvantia-en","category-alvantia-2","category-technology","tag-alvantia-en","tag-machine-learning-2","tag-software-en","not-single"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/posts\/3693","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/comments?post=3693"}],"version-history":[{"count":6,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/posts\/3693\/revisions"}],"predecessor-version":[{"id":3699,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/posts\/3693\/revisions\/3699"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/media\/3603"}],"wp:attachment":[{"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/media?parent=3693"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/categories?post=3693"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.alvantia.com\/en\/wp-json\/wp\/v2\/tags?post=3693"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}