{"id":2937,"date":"2021-04-01T10:52:41","date_gmt":"2021-04-01T10:52:41","guid":{"rendered":"http:\/\/beeeye.com\/?p=2937"},"modified":"2022-01-31T10:32:06","modified_gmt":"2022-01-31T10:32:06","slug":"assembly-line-for-predictive-models","status":"publish","type":"post","link":"http:\/\/beeeye.com\/assembly-line-for-predictive-models\/","title":{"rendered":"Assembly line for predictive models"},"content":{"rendered":"<div class=\"fusion-fullwidth fullwidth-box fusion-builder-row-1 fusion-flex-container nonhundred-percent-fullwidth non-hundred-percent-height-scrolling\" style=\"background-color: rgba(255,255,255,0);background-position: center center;background-repeat: no-repeat;border-width: 0px 0px 0px 0px;border-color:#eae9e9;border-style:solid;\" ><div class=\"fusion-builder-row fusion-row fusion-flex-align-items-flex-start\" style=\"max-width:105%;margin-left: calc(-5% \/ 2 );margin-right: calc(-5% \/ 2 );\"><div class=\"fusion-layout-column fusion_builder_column fusion-builder-column-0 fusion_builder_column_1_1 1_1 fusion-flex-column\"><div class=\"fusion-column-wrapper fusion-flex-justify-content-flex-start fusion-content-layout-column\" style=\"background-position:left top;background-repeat:no-repeat;-webkit-background-size:cover;-moz-background-size:cover;-o-background-size:cover;background-size:cover;padding: 0px 0px 0px 0px;\"><div class=\"fusion-text fusion-text-1\"><\/p>\n<p><span style=\"font-weight: 400;\">Predictive models in banks, lenders, insurance companies and others are notorious for being extremely slow to produce and launch. This is the result of many reasons, none of them directly related to the technical details of building a predictive model. Some of the reasons are:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\"><span style=\"font-weight: 400;\">Absence of direct linkage (code) between research model and the deployed model (often written in different languages!)<\/span><\/li>\n<li style=\"font-weight: 400;\"><span style=\"font-weight: 400;\">Absence of a unified data lake holding both the training data as well as the new data coming from production.<\/span><\/li>\n<li style=\"font-weight: 400;\"><span style=\"font-weight: 400;\">The need for full data lineage detailing the source and usage of every piece of data<\/span><\/li>\n<li style=\"font-weight: 400;\"><span style=\"font-weight: 400;\">The need to go through a set of regulatory procedures and approvals<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">These reasons (and more) make the modeling effort in these organizations a one-off project where each iteration or versions of the same model feels like a whole new modelling project. This in turn leads to shockingly slow modelling iterations. How slow? Some organisations require as much as six months to build and sometimes even much longer to launch a new model!<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Times like this, when a world pandemic is warping the financial universe as we know it, create an enormous stress on the modelling teams. They are required to tweak, validate and launch models in an ever increasing speed and agility. This is the enormous challenge they are facing.<\/span><\/p>\n<h2 data-fontsize=\"36\" data-lineheight=\"43.2px\" class=\"fusion-responsive-typography-calculated\" style=\"--fontSize:36; line-height: 1.2;\"><strong>Iterating on models<\/strong><\/h2>\n<p><span style=\"font-weight: 400;\">A model should be as lively and as fresh as any other production scale application code. Yes, it\u2019s better to look at a model as a code project and learn from decades of experience gained in the software industry.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A model in production should be a direct product of the model assembly line. This means that if the model (function) disappears magically, the team can reproduce the exact same model in a few seconds with the click of a button. And the line continues backwards: the same model can be re-examined in the research environment without any additional fuss. New feature engineering ideas can be ideated and tested in production in minutes, not months! Even going all the way back to the data sources used for modelling &#8211; it should be darn easy to remove or add new data points and test the outcome without too much trouble.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Such iterations are simply impossible today in the reality of a heterogeneous modelling environment. Often, the pipeline is made up of five or even more different tools which does not necessarily talk to each other.<\/span><\/p>\n<h2 data-fontsize=\"36\" data-lineheight=\"43.2px\" class=\"fusion-responsive-typography-calculated\" style=\"--fontSize:36; line-height: 1.2;\"><strong>The end to end modelling experience<\/strong><\/h2>\n<p>When models live in an end to end modelling environment, iterations measured in minutes and hours, become a reality. The entire stack needed to build, validate and deploy a model is centered around a single point of knowledge.<\/p>\n<p>We can take this even further and close the loop between the \u201cends\u201d. This creates a feedback loop which translates the lessons learnt from production data, automatically fed to the system as a new data source. Imagine that! a Shadow Modelling process running continuously and effortlessly in the background providing you with a current view of the potential of new models against the reality of production data.<\/p>\n<h2 data-fontsize=\"36\" data-lineheight=\"43.2px\" class=\"fusion-responsive-typography-calculated\" style=\"--fontSize:36; line-height: 1.2;\"><strong>Shadow Modelling<\/strong><\/h2>\n<p>The EyeOnRisk platform is an ML -based credit risk focused, end to end platform for smarter, faster risk model management. It is specifically designed to boost credit risk management processes and is built with the bank\u2019s processes and compliance needs in mind.<br>The platform encapsulates the model data flow. This lets you transfer models from research to production at an unparalleled speed. At the same time, this encapsulation allows for easier external data enrichment. Close monitoring of the data sources feeding the model while tracking data skews and suggests continual improvements.<br><a href=\"http:\/\/beeeye.com\/contact\">Contact us<\/a> to learn more<\/p>\n<h2 data-fontsize=\"36\" data-lineheight=\"43.2px\" class=\"fusion-responsive-typography-calculated\" style=\"--fontSize:36; line-height: 1.2;\"><strong>About the EyeOnRisk platform<\/strong><\/h2>\n<p>The EyeOnRisk platform is an ML -based credit risk focused, end to end platform for smarter, faster risk model management. It is specifically designed to boost credit risk management processes and is built with the bank\u2019s processes and compliance needs in mind. <\/p>\n<p>The platform encapsulates the model data flow. This lets you transfer models from research to production at an unparalleled speed. At the same time, this encapsulation allows for easier external data enrichment. Close monitoring of the data sources feeding the model while tracking data skews and suggests continual improvements.<\/p>\n<p><a href=\"http:\/\/beeeye.com\/contact\/\">Contact us <\/a>to learn more<\/p>\n<\/div><\/div><style type=\"text\/css\">.fusion-body .fusion-builder-column-0{width:100% !important;margin-top : 0px;margin-bottom : 0px;}.fusion-builder-column-0 > .fusion-column-wrapper {padding-top : 0px !important;padding-right : 0px !important;margin-right : 2.375%;padding-bottom : 0px !important;padding-left : 0px !important;margin-left : 2.375%;}@media only screen and (max-width:1024px) {.fusion-body .fusion-builder-column-0{width:100% !important;}.fusion-builder-column-0 > .fusion-column-wrapper {margin-right : 2.375%;margin-left : 2.375%;}}@media only screen and (max-width:640px) {.fusion-body .fusion-builder-column-0{width:100% !important;}.fusion-builder-column-0 > .fusion-column-wrapper {margin-right : 2.375%;margin-left : 2.375%;}}<\/style><\/div><\/div><style type=\"text\/css\">.fusion-body .fusion-flex-container.fusion-builder-row-1{ padding-top : 0px;margin-top : 0px;padding-right : 30px;padding-bottom : 0px;margin-bottom : 0px;padding-left : 30px;}<\/style><\/div>","protected":false},"excerpt":{"rendered":"","protected":false},"author":4,"featured_media":3073,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_mi_skip_tracking":false,"footnotes":""},"categories":[12,22],"tags":[17,40],"_links":{"self":[{"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/posts\/2937"}],"collection":[{"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/comments?post=2937"}],"version-history":[{"count":2,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/posts\/2937\/revisions"}],"predecessor-version":[{"id":3391,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/posts\/2937\/revisions\/3391"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/media\/3073"}],"wp:attachment":[{"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/media?parent=2937"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/categories?post=2937"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/beeeye.com\/wp-json\/wp\/v2\/tags?post=2937"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}