{"id":82060,"date":"2021-07-22T08:59:20","date_gmt":"2021-07-22T07:59:20","guid":{"rendered":"https:\/\/datascientest.com\/?p=82060"},"modified":"2023-11-09T14:06:52","modified_gmt":"2023-11-09T14:06:52","slug":"pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","status":"publish","type":"post","link":"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","title":{"rendered":"PyKana &#8211; L\u2019application de reconnaissance de  caract\u00e8res japonais)"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-post\" data-elementor-id=\"82060\" class=\"elementor elementor-82060\" data-elementor-post-type=\"post\">\n\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-79bef0b e-flex e-con-boxed e-con\" data-id=\"79bef0b\" data-element_type=\"container\" data-settings=\"{&quot;content_width&quot;:&quot;boxed&quot;}\">\n\t\t\t\t\t<div class=\"e-con-inner\">\n\t\t\t\t<div class=\"elementor-element elementor-element-43d5689 elementor-widget elementor-widget-heading\" data-id=\"43d5689\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.15.0 - 02-08-2023 *\/\n.elementor-heading-title{padding:0;margin:0;line-height:1}.elementor-widget-heading .elementor-heading-title[class*=elementor-size-]>a{color:inherit;font-size:inherit;line-height:inherit}.elementor-widget-heading .elementor-heading-title.elementor-size-small{font-size:15px}.elementor-widget-heading .elementor-heading-title.elementor-size-medium{font-size:19px}.elementor-widget-heading .elementor-heading-title.elementor-size-large{font-size:29px}.elementor-widget-heading .elementor-heading-title.elementor-size-xl{font-size:39px}.elementor-widget-heading .elementor-heading-title.elementor-size-xxl{font-size:59px}<\/style><h2 class=\"elementor-heading-title elementor-size-default\">Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021. Leur objectif  est d\u2019utiliser diff\u00e9rents algorithmes de Machine et Deep Learning afin de r\u00e9aliser de la reconnaissance de Kanas, des caract\u00e8res japonais. <\/h2>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t<div class=\"elementor-element elementor-element-4657dae e-flex e-con-boxed e-con\" data-id=\"4657dae\" data-element_type=\"container\" data-settings=\"{&quot;content_width&quot;:&quot;boxed&quot;}\">\n\t\t\t\t\t<div class=\"e-con-inner\">\n\t\t\t\t<div class=\"elementor-element elementor-element-f4e0e94 elementor-widget elementor-widget-heading\" data-id=\"f4e0e94\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h3 class=\"elementor-heading-title elementor-size-default\">Comment ont-ils eu l\u2019id\u00e9e du projet\u202f? Ils vous le racontent.<\/h3>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-79fd2c3 elementor-widget elementor-widget-text-editor\" data-id=\"79fd2c3\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.15.0 - 02-08-2023 *\/\n.elementor-widget-text-editor.elementor-drop-cap-view-stacked .elementor-drop-cap{background-color:#69727d;color:#fff}.elementor-widget-text-editor.elementor-drop-cap-view-framed .elementor-drop-cap{color:#69727d;border:3px solid;background-color:transparent}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap{margin-top:8px}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap-letter{width:1em;height:1em}.elementor-widget-text-editor .elementor-drop-cap{float:left;text-align:center;line-height:1;font-size:50px}.elementor-widget-text-editor .elementor-drop-cap-letter{display:inline-block}<\/style>\t\t\t\t<p><span style=\"font-weight: 400;\">\u201cNous partageons un int\u00e9r\u00eat commun pour la culture japonaise. Par ailleurs, Dahiana une des membres de notre \u00e9quipe r\u00e9sidait au Japon. Elle a jou\u00e9 un r\u00f4le essentiel au niveau de l\u2019apprentissage des bases des diff\u00e9rents syst\u00e8mes d\u2019\u00e9criture japonais. Nous avions eu de la chance de la compter parmi les membres de notre \u00e9quipe. D\u2019autre part, certains d\u2019entre nous ont d\u00e9j\u00e0 travaill\u00e9 sur des th\u00e9matiques de traitement d\u2019image. Ainsi, l\u2019attrait pour le projet s\u2019est fait naturellement.\u00a0\u00a0<\/span><\/p><blockquote><p><strong>De plus, la r\u00e9alisation de ce projet laisse entrevoir des applications concr\u00e8tes dans le r\u00e9el\u00a0: la reconnaissance de caract\u00e8res (ici japonais, mais peut \u00eatre g\u00e9n\u00e9ralis\u00e9e \u00e0 d\u2019autres) aboutissant \u00e0 de la traduction en temps r\u00e9el ou de la lecture de fichiers manuscrits num\u00e9ris\u00e9s. Bien que ces applications existent d\u00e9j\u00e0, il est toujours valorisant de travailler sur un projet dont la port\u00e9e est palpable.<\/strong><\/p><\/blockquote><p><span style=\"font-weight: 400;\">Enfin, ce projet nous a essentiellement permis de mettre en pratique ce que nous avions \u00e9tudi\u00e9 au cours de notre formation chez Datascientest comme la data visualisation, le machine learning, le deep learning ainsi que les applications web.\u00a0\u00bb<\/span><\/p><p><span style=\"font-weight: 400;\">Les codes d\u00e9velopp\u00e9s lors de ce projet se trouvent dans ce <a href=\"https:\/\/github.com\/DataScientest\/\">repository GitHub<\/a>.<\/span><\/p><p><span style=\"font-weight: 400;\">Vous pouvez retrouver la d\u00e9mo streamlit compl\u00e8te sur le site du\u00a0<\/span><a href=\"https:\/\/studio.datascientest.com\/project\/pykana\/\" target=\"_blank\" rel=\"noopener\">Studio de DataScientest<\/a><span style=\"font-weight: 400;\">\u00a0.\u00a0<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-b1e1749 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"b1e1749\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-e4390e0\" data-id=\"e4390e0\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-2ccc4c3 elementor-widget elementor-widget-heading\" data-id=\"2ccc4c3\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h2 class=\"elementor-heading-title elementor-size-default\">Projet PyKana<\/h2>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-8bc699b elementor-widget elementor-widget-toggle\" data-id=\"8bc699b\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>.elementor-toggle{text-align:left}.elementor-toggle .elementor-tab-title{font-weight:700;line-height:1;margin:0;padding:15px;border-bottom:1px solid #d5d8dc;cursor:pointer;outline:none}.elementor-toggle .elementor-tab-title .elementor-toggle-icon{display:inline-block;width:1em}.elementor-toggle .elementor-tab-title .elementor-toggle-icon svg{-webkit-margin-start:-5px;margin-inline-start:-5px;width:1em;height:1em}.elementor-toggle .elementor-tab-title .elementor-toggle-icon.elementor-toggle-icon-right{float:right;text-align:right}.elementor-toggle .elementor-tab-title .elementor-toggle-icon.elementor-toggle-icon-left{float:left;text-align:left}.elementor-toggle .elementor-tab-title .elementor-toggle-icon .elementor-toggle-icon-closed{display:block}.elementor-toggle .elementor-tab-title .elementor-toggle-icon .elementor-toggle-icon-opened{display:none}.elementor-toggle .elementor-tab-title.elementor-active{border-bottom:none}.elementor-toggle .elementor-tab-title.elementor-active .elementor-toggle-icon-closed{display:none}.elementor-toggle .elementor-tab-title.elementor-active .elementor-toggle-icon-opened{display:block}.elementor-toggle .elementor-tab-content{padding:15px;border-bottom:1px solid #d5d8dc;display:none}@media (max-width:767px){.elementor-toggle .elementor-tab-title{padding:12px}.elementor-toggle .elementor-tab-content{padding:12px 10px}}.e-con-inner>.elementor-widget-toggle,.e-con>.elementor-widget-toggle{width:var(--container-widget-width);--flex-grow:var(--container-widget-flex-grow)}<\/style>\t\t<div class=\"elementor-toggle\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1461\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"button\" aria-controls=\"elementor-tab-content-1461\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">1. Data Exploration<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1461\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"region\" aria-labelledby=\"elementor-tab-title-1461\"><p><span style=\"font-weight: 400; color: #000000;\">Les caract\u00e8res japonais sont principalement r\u00e9partis en 2 cat\u00e9gories :<\/span><\/p><ul><li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"color: #000000;\"><b>Kanji<\/b><span style=\"font-weight: 400;\"> : d\u00e9riv\u00e9s du chinois et servent \u00e0 lire et \u00e9crire une partie du japonais en \u00e9tant associ\u00e9s \u00e0 des kanas<\/span><\/span><ul><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Les kanjis comportent entre 2000 et 4000 caract\u00e8res<\/span><\/li><\/ul><\/li><li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"color: #000000;\"><b>Kana<\/b><span style=\"font-weight: 400;\"> : servent \u00e0 noter phon\u00e9tiquement la langue japonaise et sont constitu\u00e9s de 2 sous-groupes<\/span><\/span><ul><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"color: #000000;\"><b><i>Hiragana<\/i><\/b><span style=\"font-weight: 400;\"> : 48 caract\u00e8res<\/span><\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"color: #000000;\"><b><i>Katakana<\/i><\/b><span style=\"font-weight: 400;\"> : 48 caract\u00e8res<\/span><\/span><\/li><\/ul><\/li><\/ul><p><span style=\"color: #000000;\"><span style=\"font-weight: 400;\">\u00c9tant dans un probl\u00e8me d\u2019apprentissage supervis\u00e9 et au regard de la puissance de calcul que nous avions \u00e0 disposition, nous nous sommes limit\u00e9s \u00e0 la reconnaissance des <\/span><b>kana<\/b><span style=\"font-weight: 400;\">.<\/span><\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1462\" class=\"elementor-tab-title\" data-tab=\"2\" role=\"button\" aria-controls=\"elementor-tab-content-1462\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">2. Data Visualisation<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1462\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"2\" role=\"region\" aria-labelledby=\"elementor-tab-title-1462\"><p><span style=\"font-weight: 400; color: #000000;\">Pour les besoins de notre \u00e9tude, nous nous sommes bas\u00e9s sur le jeu de donn\u00e9es du site\u00a0<\/span><a href=\"http:\/\/etlcdb.db.aist.go.jp\/\" target=\"_blank\" rel=\"noopener\">ETL Character Database<\/a><span style=\"font-weight: 400; color: #000000;\">\u00a0qui nous fournissait des images de kanas compress\u00e9es, toutes en niveaux de gris.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Nous avions donc \u00e0 disposition environ 6100 Hiragana et 10600 Katakana \u00e9crits respectivement par 120 et 106 personnes qui devaient \u00e9crire 2 ou 3 fois chaque kana.<\/span><\/p><p><img decoding=\"async\" class=\"wp-image-82062 size-medium aligncenter\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/OK1HftidkOG0Au5rz2_i9-HH47AbkatKD7rDgbzA1SkssVmjdvGHydolZTnirlVoZ1ELszEiCUmUz-J34b6Amn3qWwVQsEtq_flaAaKQe-EXBTzUFECAuNTRBRa37i4vi-300x260.jpg\" alt=\"extrait de la base de donn\u00e9es Hiragana\" width=\"300\" height=\"260\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/OK1HftidkOG0Au5rz2_i9-HH47AbkatKD7rDgbzA1SkssVmjdvGHydolZTnirlVoZ1ELszEiCUmUz-J34b6Amn3qWwVQsEtq_flaAaKQe-EXBTzUFECAuNTRBRa37i4vi-300x260.jpg 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/OK1HftidkOG0Au5rz2_i9-HH47AbkatKD7rDgbzA1SkssVmjdvGHydolZTnirlVoZ1ELszEiCUmUz-J34b6Amn3qWwVQsEtq_flaAaKQe-EXBTzUFECAuNTRBRa37i4vi-768x665.jpg 768w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/OK1HftidkOG0Au5rz2_i9-HH47AbkatKD7rDgbzA1SkssVmjdvGHydolZTnirlVoZ1ELszEiCUmUz-J34b6Amn3qWwVQsEtq_flaAaKQe-EXBTzUFECAuNTRBRa37i4vi.jpg 966w\" sizes=\"(max-width: 300px) 100vw, 300px\" \/><br \/><span style=\"text-decoration: underline;\">Fig. 2.1 Extrait de la base de donn\u00e9es brutes Hiragana<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Les caract\u00e8res de la base \u00e9taient <\/span><span style=\"color: #000000;\"><b>concat\u00e9n\u00e9s dans des images que nous avons d\u00fb s\u00e9parer<\/b><\/span><span style=\"font-weight: 400; color: #000000;\">, ainsi nous avons pu faire le travail de reconnaissance sur chaque kana individuellement.<\/span><\/p><p><span style=\"color: #000000;\"><span style=\"font-weight: 400;\">Apr\u00e8s observation d\u00e9taill\u00e9e<\/span><span style=\"font-size: 15px; font-weight: 400;\">\u00a0des bases de donn\u00e9es brutes des Hiragana et Katakana, nous avons remarqu\u00e9 que notre set d\u2019\u00e9tude \u00e9tait assez bruit\u00e9, comme nous pouvo<\/span><span style=\"font-size: 15px; font-weight: 400;\">ns le vo<\/span><\/span><span style=\"color: #000000; font-size: 15px; font-weight: 400;\">ir ci-dessous :<\/span><\/p><p><img decoding=\"async\" class=\"wp-image-82078 size-medium aligncenter\" style=\"color: var( --e-global-color-text ); font-family: Metropolis, sans-serif; font-size: 16px; font-style: normal; font-variant-ligatures: normal; font-variant-caps: normal; font-weight: 400;\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-300x139.jpg\" alt=\"Fig. 2.2 Extraits de Hiragana bruts\" width=\"300\" height=\"139\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-300x139.jpg 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1024x476.jpg 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-768x357.jpg 768w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO.jpg 1223w\" sizes=\"(max-width: 300px) 100vw, 300px\" \/><br \/><span style=\"text-decoration: underline;\">Fig. 2.2 Extraits de Hiragana bruts<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Initialement nous pensions que le bruit pourrait biaiser notre \u00e9tude, donc nous avons appliqu\u00e9 nos algorithmes sur nos donn\u00e9es brutes et filtr\u00e9es afin d\u2019\u00e9valuer ce biais. Ci-dessous, voici des extraits de nos kanas filtr\u00e9s :<\/span><\/p><p><img decoding=\"async\" class=\"size-medium wp-image-82079 aligncenter\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1-300x139.jpg\" alt=\"Fig. 2.3 Extraits de Katakana filtr\u00e9s\" width=\"300\" height=\"139\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1-300x139.jpg 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1-1024x476.jpg 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1-768x357.jpg 768w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/rqURfOOIjRBxttMtW1C6RaGoHWi9dR1ke31xuUWok5R9Ots6HOE3o2bF6ObT24khyagnbbTZIzq3-RKHok2z155bl32YkWfcuK2Qz75LjBihspA3qyRAk5uI9w9uGSEgO-1.jpg 1223w\" sizes=\"(max-width: 300px) 100vw, 300px\" \/> <br \/><span style=\"text-decoration: underline;\">Fig. 2.3 Extraits de Katakana filtr\u00e9s<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Ci-dessous, nous pouvons voir que notre jeu de donn\u00e9es \u00e9tait assez \u00e9quilibr\u00e9 pour les Hiragana et Katakana :<\/span><\/p><p><img decoding=\"async\" class=\"size-medium wp-image-82080 aligncenter\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/2JBq2j-wThWdCDJ8XrEfxG5ghcu-v5vTMsYLfkn7X7p2T19XZLTXp1c1GQSDoBMMXY5vyj0QdzjtqRdPtzq0FgPCzJmhP5IMhpcG0VE9m_BTSnZpsVLAR3DbDbh9fkCdp-290x300.jpg\" alt=\"Distribution des Hiragana et Katakana dans la base d\u2019images\" width=\"290\" height=\"300\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/2JBq2j-wThWdCDJ8XrEfxG5ghcu-v5vTMsYLfkn7X7p2T19XZLTXp1c1GQSDoBMMXY5vyj0QdzjtqRdPtzq0FgPCzJmhP5IMhpcG0VE9m_BTSnZpsVLAR3DbDbh9fkCdp-290x300.jpg 290w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/2JBq2j-wThWdCDJ8XrEfxG5ghcu-v5vTMsYLfkn7X7p2T19XZLTXp1c1GQSDoBMMXY5vyj0QdzjtqRdPtzq0FgPCzJmhP5IMhpcG0VE9m_BTSnZpsVLAR3DbDbh9fkCdp.jpg 503w\" sizes=\"(max-width: 290px) 100vw, 290px\" \/><\/p><p><img decoding=\"async\" class=\"size-medium wp-image-82081 aligncenter\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/G3f42MN5T2XpdeL5eKfqakw54Zs5Xh1U3RHfu5vukNDcfoFbSFxwlVShejT8SIQKgZSOKYEFyh8-iKpLPPvBK7lZNFfS52jEEk3zdpTNt_NjAOw6Pltv1lcDj44dWHCWy-290x300.jpg\" alt=\"Distribution des Hiragana et Katakana dans la base d\u2019images\" width=\"290\" height=\"300\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/G3f42MN5T2XpdeL5eKfqakw54Zs5Xh1U3RHfu5vukNDcfoFbSFxwlVShejT8SIQKgZSOKYEFyh8-iKpLPPvBK7lZNFfS52jEEk3zdpTNt_NjAOw6Pltv1lcDj44dWHCWy-290x300.jpg 290w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/G3f42MN5T2XpdeL5eKfqakw54Zs5Xh1U3RHfu5vukNDcfoFbSFxwlVShejT8SIQKgZSOKYEFyh8-iKpLPPvBK7lZNFfS52jEEk3zdpTNt_NjAOw6Pltv1lcDj44dWHCWy.jpg 503w\" sizes=\"(max-width: 290px) 100vw, 290px\" \/> <br \/><span style=\"text-decoration: underline;\">Fig. 2.4 Distribution des Hiragana et Katakana dans la base d\u2019images<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1463\" class=\"elementor-tab-title\" data-tab=\"3\" role=\"button\" aria-controls=\"elementor-tab-content-1463\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">3. Machine Learning<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1463\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"3\" role=\"region\" aria-labelledby=\"elementor-tab-title-1463\"><p style=\"text-align: left;\"><span style=\"font-weight: 400; color: #000000;\">Nous avons retenu <\/span><span style=\"color: #000000;\"><b>4 mod\u00e8les classiques de Machine Learning<\/b><\/span><span style=\"font-weight: 400; color: #000000;\">, conjugu\u00e9s d\u2019un Voting Classifier afin de r\u00e9soudre notre probl\u00e8me de classification :<\/span><b><\/b><\/p><ul><li aria-level=\"1\"><a href=\"https:\/\/datascientest.com\/random-forest-definition\" target=\"_blank\" rel=\"noopener\">Random Forest<\/a><\/li><\/ul><ul><li aria-level=\"1\"><a href=\"https:\/\/datascientest.com\/knn\" target=\"_blank\" rel=\"noopener\">KNN (K-Nearest Neighbors ou K plus proches voisins)<\/a><\/li><\/ul><ul><li aria-level=\"1\"><a href=\"https:\/\/datascientest.com\/svm\" target=\"_blank\" rel=\"noopener\">SVM (Support Vector Machine)<\/a><br \/><p>\u00a0<\/p><\/li><li aria-level=\"1\"><a href=\"https:\/\/datascientest.com\/regression-logistique-quest-ce-que-cest\" target=\"_blank\" rel=\"noopener\">Logistic Regression<\/a><\/li><\/ul><p><span style=\"color: #000000;\"><span style=\"font-weight: 400;\">Nous avons tent\u00e9 d\u2019optimiser ces mod\u00e8les ind\u00e9pendamment les uns des autres et r\u00e9aliser ce travail sur nos donn\u00e9es brutes et filtr\u00e9es avec <\/span><b>Canny<\/b><span style=\"font-weight: 400;\"> (extracteur de contours), ainsi avons-nous pu voir si la r\u00e9duction du bruit en entr\u00e9e influe sur les performances des mod\u00e8les. Puis nous avons combin\u00e9 chaque mod\u00e8le avec ses meilleurs param\u00e8tres pour le <\/span><b>Voting Classifier<\/b><span style=\"font-weight: 400;\">.\u00a0<\/span><\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Nous avons utilis\u00e9 80% des donn\u00e9es pour l\u2019entra\u00eenement, et 20% des donn\u00e9es restantes pour le test. Afin de pouvoir comparer les performances des divers mod\u00e8les, nous avons fix\u00e9 ces ensembles d\u2019entra\u00eenement et de test.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Enfin, nous avons obtenu les scores de pr\u00e9dictions ci-dessous :<\/span><img decoding=\"async\" class=\"alignnone wp-image-82091 size-full\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr.jpg\" alt=\"les scores de pr\u00e9dictions\" width=\"1600\" height=\"900\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr.jpg 1600w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr-300x169.jpg 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr-1024x576.jpg 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr-768x432.jpg 768w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Logistic-Regr-1536x864.jpg 1536w\" sizes=\"(max-width: 1600px) 100vw, 1600px\" \/><br \/><u>Les scores de pr\u00e9dictions<\/u><\/p><p>\u00a0<\/p><p><span style=\"font-weight: 400; color: #000000;\">La premi\u00e8re remarque int\u00e9ressante que nous pouvons faire est que contrairement \u00e0 l\u2019hypoth\u00e8se que nous avions formul\u00e9e au d\u00e9but, le filtrage de nos images d\u00e9grade la qualit\u00e9 de nos pr\u00e9dictions.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Ensuite, le mod\u00e8le SVM semble donner les meilleurs r\u00e9sultats en termes de pr\u00e9diction, ce qui est coh\u00e9rent avec le fait qu\u2019il soit r\u00e9put\u00e9 performant pour les probl\u00e9matiques de classification d\u2019image. Nous avons ensuite utilis\u00e9 le Voting Classifier pour tenter d\u2019optimiser les r\u00e9sultats. Cependant, le gain de score obtenu est trop faible par rapport au temps de calcul n\u00e9cessaire. Si l\u2019on reste \u00e0 ce niveau-l\u00e0 d\u2019\u00e9tude, les mod\u00e8les les plus efficaces de Machine Learning \u00e0 utiliser sont le <\/span><span style=\"color: #000000;\"><b>Random Forest ou le SVM<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> pour la reconnaissance de kanas.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Enfin, les <\/span><span style=\"color: #000000;\"><b>Katakana semblent plus faciles \u00e0 pr\u00e9dire que les Hiragana<\/b><\/span><span style=\"font-weight: 400; color: #000000;\">. En s\u2019int\u00e9ressant plus pr\u00e9cis\u00e9ment \u00e0 la forme de ces kanas, cela semble logique, car la forme des Hiragana est bien plus \u201carrondie\u201d et complexe que celle des Katakana (cf. figures 2.2, 2.3).<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1464\" class=\"elementor-tab-title\" data-tab=\"4\" role=\"button\" aria-controls=\"elementor-tab-content-1464\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">4. Deep Learning<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1464\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"4\" role=\"region\" aria-labelledby=\"elementor-tab-title-1464\"><p><span style=\"font-weight: 400; color: #000000;\">Devant les limites observ\u00e9es par les algorithmes de Machine Learning pour notre \u00e9tude, nous avons embray\u00e9 sur du Deep Learning avec des r\u00e9seaux de neurones\u00a0<\/span><a href=\"https:\/\/datascientest.com\/convolutional-neural-network\" target=\"_blank\" rel=\"noopener\">CNN (Convolutional Neural Network)<\/a><span style=\"font-weight: 400; color: #000000;\">.<\/span><\/p><ol><li style=\"font-weight: 400;\" aria-level=\"3\"><h2><strong><span style=\"color: #000000;\">Mod\u00e8le CNN utilis\u00e9<\/span><\/strong><\/h2><\/li><\/ol><p><span style=\"color: #000000;\"><span style=\"font-weight: 400;\">Nous avons d\u00e9velopp\u00e9 l\u2019architecture CNN ci-apr\u00e8s (<\/span><i><span style=\"font-weight: 400;\">Fig. 2.5<\/span><\/i><span style=\"font-weight: 400;\">) :<\/span><\/span><img decoding=\"async\" class=\"aligncenter wp-image-82100 size-full\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0.jpg\" alt=\"Architecture du mod\u00e8le CNN utilis\u00e9 \" width=\"1600\" height=\"900\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0.jpg 1600w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0-300x169.jpg 300w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0-1024x576.jpg 1024w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0-768x432.jpg 768w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/qzkEnC8cM2ahpDoDeq-IchsEOtpkQFMX2FUFkVMzR-fVcAHc4TSjD4qoK36G78dnmmdv1isaEL41muKOTLC6Q3dh8j4Euq3saW79oebpeKWGqc9dC3RkvVjgNQqbFBBd0-1536x864.jpg 1536w\" sizes=\"(max-width: 1600px) 100vw, 1600px\" \/><br \/><span style=\"text-decoration: underline;\">Fig. 2.5 : Architecture du mod\u00e8le CNN utilis\u00e9<\/span><\/p><p><span style=\"color: #000000;\"><span style=\"font-weight: 400;\">Ci-apr\u00e8s le d\u00e9tail des couches constituant le mod\u00e8le CNN utilis\u00e9 (<\/span><i><span style=\"font-weight: 400;\">Fig. 2.6<\/span><\/i><span style=\"font-weight: 400;\">) :<br \/><\/span><\/span><\/p><p><img decoding=\"async\" class=\"aligncenter wp-image-82101 size-full\" src=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Layer-type.jpg\" alt=\"R\u00e9sum\u00e9 du mod\u00e8le CNN utilis\u00e9\" width=\"589\" height=\"511\" srcset=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Layer-type.jpg 589w, https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/Layer-type-300x260.jpg 300w\" sizes=\"(max-width: 589px) 100vw, 589px\" \/><br \/><span style=\"text-decoration: underline;\">\u00a0Fig. 2.6 R\u00e9sum\u00e9 du mod\u00e8le CNN utilis\u00e9<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Avec ce mod\u00e8le CNN, nous obtenions :<br \/><\/span><\/p><table style=\"height: 105px;\" width=\"254\"><tbody><tr><td><b>M\u00e9trique<\/b><\/td><td><b>Hiragana<\/b><\/td><td><b>Katakana<\/b><\/td><\/tr><tr><td><span style=\"font-weight: 400;\">Score<\/span><\/td><td><span style=\"font-weight: 400;\">89%<\/span><\/td><td><span style=\"font-weight: 400;\">99%<\/span><\/td><\/tr><\/tbody><\/table><p><span style=\"font-weight: 400; color: #000000;\">Pour les Hiragana, nous avons obtenu des r\u00e9sultats assez satisfaisants 89% et une courbe qui converge vers un maximum et s\u2019est stabilis\u00e9e vers cette valeur, apr\u00e8s un entra\u00eenement de 50 epochs.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Sur les Katakana, sur le jeu de donn\u00e9es de test, nous avons obtenu des r\u00e9sultats tr\u00e8s suffisants de 99% et une courbe qui converge vers un maximum et s\u2019est stabilis\u00e9e vers cette valeur, apr\u00e8s un entra\u00eenement de 50 epochs.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Nous pouvons noter que nous avons d\u00e9j\u00e0 un gros gain de score par rapport aux mod\u00e8les de Machine Learning utilis\u00e9s.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Bien que les r\u00e9sultats obtenus \u00e9taient assez satisfaisants avec le mod\u00e8le CNN utilis\u00e9, nous avons voulu optimiser cette approche en faisant de la data augmentation : ceci est un moyen permettant \u00e0 partir d&rsquo;un jeu de donn\u00e9es de multiplier les donn\u00e9es disponibles pour l&rsquo;entra\u00eenement du r\u00e9seau afin de permettre d&rsquo;avoir plus de donn\u00e9es pour r\u00e9aliser l&rsquo;entra\u00eenement et la mise \u00e0 jour des poids du r\u00e9seau.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Avec la data augmentation, nous obtenions :<\/span><\/p><table><tbody><tr><td><span style=\"color: #000000;\"><b>M\u00e9trique<\/b><\/span><\/td><td><span style=\"color: #000000;\"><b>Hiragana<\/b><\/span><\/td><td><span style=\"color: #000000;\"><b>Katakana<\/b><\/span><\/td><\/tr><tr><td><span style=\"font-weight: 400; color: #000000;\">Score<\/span><\/td><td><span style=\"font-weight: 400; color: #000000;\">95%<\/span><\/td><td><span style=\"font-weight: 400; color: #000000;\">99%<\/span><\/td><\/tr><\/tbody><\/table><p><span style=\"font-weight: 400; color: #000000;\">Ce r\u00e9seau est tr\u00e8s performant :\u00a0<\/span><\/p><ul><li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400; color: #000000;\">Sur les Katakana, un score sup\u00e9rieur \u00e0 99% pour la reconnaissance des caract\u00e8res<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400; color: #000000;\">Sur les Hiragana, un score sup\u00e9rieur \u00e0 95% pour la reconnaissance des caract\u00e8res<\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">Par manque de temps, nous n\u2019avons pas pu exploiter toutes les pistes afin d\u2019am\u00e9liorer et optimiser notre mod\u00e8le, voire faire un mod\u00e8le pour chaque type de kana, un sp\u00e9cifique pour les Hiragana et un autre pour les Katakana.<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-1465\" class=\"elementor-tab-title\" data-tab=\"5\" role=\"button\" aria-controls=\"elementor-tab-content-1465\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">5. Limites \/ Axes d\u2019am\u00e9lioration<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-1465\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"5\" role=\"region\" aria-labelledby=\"elementor-tab-title-1465\"><h2><strong>1. Mod\u00e8les de Machine Learning<\/strong><\/h2><p><span style=\"font-weight: 400; color: #000000;\">Afin d\u2019am\u00e9liorer la pr\u00e9cision des algorithmes de Machine Learning utilis\u00e9s, nous pouvons envisager :<\/span><b><\/b><\/p><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Transfer Learning<\/b><\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">Cr\u00e9ation d\u2019un mod\u00e8le hybride combinant les meilleurs mod\u00e8les de Machine Learning (SVM et Random Forest) ou Machine\/Deep Learning, notamment pour augmenter les performances sur les Hiragana<\/span><b><\/b><\/p><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Autre m\u00e9thode d\u2019\u00e9limination du bruit<\/b><\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">Nous n\u2019avons eu le temps que de tester le filtre Canny pour faire la suppression de bruit sur nos caract\u00e8res en entr\u00e9e, mais celui-ci n\u2019est peut-\u00eatre pas adapt\u00e9 \u00e0 notre probl\u00e9matique. L\u2019utilisation d\u2019un autre filtre donnerait peut-\u00eatre de meilleurs r\u00e9sultats.<\/span><b><\/b><\/p><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>\u201cG\u00e9n\u00e9ralisation\u201d du mod\u00e8le<\/b><\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">Dans le cadre de notre projet, nous n\u2019avons utilis\u00e9 que des mod\u00e8les permettant de faire de la reconnaissance sur un seul caract\u00e8re \u00e0 la fois. L\u2019utilisation de mod\u00e8les fonctionnant sur des s\u00e9quences de caract\u00e8res donnerait une dimension beaucoup plus grande \u00e0 notre projet en termes d\u2019applications.<\/span><\/p><h2><strong><span style=\"color: #000000;\">2. Mod\u00e8le de Deep Learning<\/span><\/strong><\/h2><p><span style=\"font-weight: 400; color: #000000;\">Afin d\u2019am\u00e9liorer notre mod\u00e8le CNN (d\u00e9j\u00e0 performant), nous pouvons envisager :<\/span><b><\/b><\/p><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Optimisation de l\u2019architecture r\u00e9seau<\/b><\/span><\/li><\/ul><ul><li style=\"list-style-type: none;\"><ul><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Nombre de couches de Convolutions \/ Pooling<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Param\u00e8tres associ\u00e9s aux couches de Convolutions et de Pooling (strides, filters, kernel_size&#8230;)<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Couches de Regularization (limitation de l&rsquo;over-fitting)<\/span><\/li><\/ul><\/li><\/ul><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Optimisation du traitement des donn\u00e9es<\/b><\/span><\/li><\/ul><ul><li style=\"list-style-type: none;\"><ul><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Param\u00e8tres de transformation d&rsquo;images (rotation, zoom, flip &#8230;)<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Nombre de transformations<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Nombre d&rsquo;images d&rsquo;entra\u00eenement<\/span><\/li><\/ul><\/li><\/ul><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Assemblage de mod\u00e8les<\/b><\/span><\/li><\/ul><ul><li style=\"list-style-type: none;\"><ul><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Il est possible d&rsquo;assembler plusieurs mod\u00e8les diff\u00e9rents<\/span><\/li><li style=\"font-weight: 400;\" aria-level=\"2\"><span style=\"font-weight: 400; color: #000000;\">Les pr\u00e9dictions sont alors moyenn\u00e9es entre les diff\u00e9rents mod\u00e8les<\/span><\/li><\/ul><\/li><\/ul><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>\u201cG\u00e9n\u00e9ralisation\u201d du mod\u00e8le<\/b><\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">Dans le cadre de notre projet, nous n\u2019avons utilis\u00e9 que des mod\u00e8les permettant de faire de la reconnaissance sur un seul caract\u00e8re \u00e0 la fois. L\u2019utilisation de mod\u00e8les fonctionnant sur des s\u00e9quences de caract\u00e8res donnerait une dimension beaucoup plus grande \u00e0 notre projet en termes d\u2019applications.<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-b3b0493 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"b3b0493\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-bd63a39\" data-id=\"bd63a39\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t\t\t<div class=\"elementor-element elementor-element-de92f44 elementor-widget elementor-widget-heading\" data-id=\"de92f44\" data-element_type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<h2 class=\"elementor-heading-title elementor-size-default\">D\u00e9roulement du projet<\/h2>\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-e5b98ff elementor-widget elementor-widget-toggle\" data-id=\"e5b98ff\" data-element_type=\"widget\" data-widget_type=\"toggle.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-toggle\">\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-2401\" class=\"elementor-tab-title\" data-tab=\"1\" role=\"button\" aria-controls=\"elementor-tab-content-2401\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">Difficult\u00e9s rencontr\u00e9es<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-2401\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"1\" role=\"region\" aria-labelledby=\"elementor-tab-title-2401\"><p><span style=\"font-weight: 400; color: #000000;\">La principale difficult\u00e9 que nous avons rencontr\u00e9e au d\u00e9but du projet a \u00e9t\u00e9 le traitement\/extraction de nos donn\u00e9es brutes de la base<\/span><a href=\"http:\/\/etlcdb.db.aist.go.jp\/\" target=\"_blank\" rel=\"noopener\"> ETL Character Database<\/a><span style=\"font-weight: 400; color: #000000;\">. Nous nous sommes en effet aper\u00e7us qu\u2019il y avait des probl\u00e8mes d\u2019encodage compliquant l\u2019extraction des images. Nous avons n\u00e9anmoins pu les r\u00e9cup\u00e9rer sur un de nos ordinateurs et les dupliquer sur les autres afin que tout le monde puisse travailler sur la m\u00eame base.\u00a0<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Au niveau des mod\u00e8les \u00e0 essayer nous n\u2019avons pas rencontr\u00e9 de difficult\u00e9s majeures, si ce n\u2019est que nous ne sommes pas partis directement vers les mod\u00e8les de Deep Learning,\u00a0 qui ont \u00e9t\u00e9 \u00e9tudi\u00e9s plus tard dans la formation. Cependant, cela nous a permis de voir qu\u2019il \u00e9tait possible d\u2019obtenir des r\u00e9sultats corrects avec les mod\u00e8les de Machine Learning (m\u00eame si moins bons que ceux de Deep Learning), et qu\u2019un meilleur travail de filtrage de bruit sur nos images d\u2019entr\u00e9e pourrait peut-\u00eatre am\u00e9liorer ces r\u00e9sultats.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">En termes de calendrier, nous avons mis en g\u00e9n\u00e9ral plus de temps que pr\u00e9vu \u00e0 r\u00e9cup\u00e9rer nos donn\u00e9es \u00e0 cause des probl\u00e8mes d\u2019encodage. Idem pour optimiser nos mod\u00e8les de Machine Learning, \u00e0 cause du temps d\u2019ex\u00e9cution trop long pour les Grid Search (recherche des meilleurs param\u00e8tres pour chaque mod\u00e8le). Cela a eu un impact notamment sur la fin de notre projet o\u00f9 nous aurions aim\u00e9 pouvoir peut-\u00eatre mettre en place un mod\u00e8le de\u00a0<\/span><a href=\"https:\/\/datascientest.com\/transfer-learning\" target=\"_blank\" rel=\"noopener\">Transfer Learning<\/a><span style=\"font-weight: 400; color: #000000;\">, ou un mod\u00e8le hybride entre le\u00a0<\/span><a href=\"https:\/\/datascientest.com\/machine-learning-tout-savoir\" target=\"_blank\" rel=\"noopener\">Machine Learning<\/a><span style=\"font-weight: 400; color: #000000;\">\u00a0et le\u00a0<\/span><a href=\"https:\/\/datascientest.com\/deep-learning\" target=\"_blank\" rel=\"noopener\">Deep Learning<\/a><span style=\"font-weight: 400; color: #000000;\">, notamment pour am\u00e9liorer nos r\u00e9sultats sur les Hiragana.\u00a0<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Comme d\u00e9crit pr\u00e9c\u00e9demment, nous avons rencontr\u00e9 au d\u00e9part des probl\u00e8mes de lecture de nos donn\u00e9es \u00e0 cause de l\u2019encodage. Puis pour g\u00e9rer la volum\u00e9trie des images individuelles, nous avons \u00e9galement fait le choix de <\/span><span style=\"color: #000000;\"><b>transformer les datas pour les stocker dans des fichiers CSV plus simples \u00e0 importer<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> \u00e0 chaque fois et moins co\u00fbteux en termes de m\u00e9moire.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Enfin, surtout au d\u00e9but nous pensions avoir un probl\u00e8me de bruit sur nos images qui nuirait \u00e0 la performance de nos mod\u00e8les, ce qui s\u2019est r\u00e9v\u00e9l\u00e9 faux par la suite, mais nous a fait perdre un temps certain<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-2402\" class=\"elementor-tab-title\" data-tab=\"2\" role=\"button\" aria-controls=\"elementor-tab-content-2402\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">Accompagnement<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-2402\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"2\" role=\"region\" aria-labelledby=\"elementor-tab-title-2402\"><p><span style=\"font-weight: 400; color: #000000;\">En plus d\u2019avoir pu b\u00e9n\u00e9ficier des masterclass dispens\u00e9es par DataScientest, nous avions \u00e0 disposition les diff\u00e9rents modules en ligne sur la plateforme pour nous aider \u00e0 d\u00e9velopper nos diff\u00e9rents algorithmes. DataScientest nous a aussi propos\u00e9 un calendrier de d\u00e9roulement du projet, sur lequel nous nous sommes appuy\u00e9s pour travailler en m\u00e9thode agile.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Pendant notre projet, nous avons \u00e9galement \u00e9t\u00e9 encadr\u00e9s par Louis qui nous a \u00e9t\u00e9 par son exp\u00e9rience de tr\u00e8s bon conseil en ce qui concerne le d\u00e9roulement des \u00e9tapes du projet :<\/span><\/p><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Outils \u00e0 utiliser pour r\u00e9aliser la data visualisation<\/b><\/span><\/li><\/ul><ul><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Algorithmes les plus pertinents \u00e0 utiliser pour notre \u00e9tude<\/b><\/span><\/li><li aria-level=\"1\"><span style=\"color: #000000;\"><b>Validation de la coh\u00e9rence des r\u00e9sultats obtenus<\/b><\/span><\/li><\/ul><p><span style=\"font-weight: 400; color: #000000;\">De plus, le regard de Louis nous permettait de prendre un peu de hauteur sur nos travaux afin qu\u2019ils soient intelligibles pour les personnes ext\u00e9rieures au projet.<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<div class=\"elementor-toggle-item\">\n\t\t\t\t\t<div id=\"elementor-tab-title-2403\" class=\"elementor-tab-title\" data-tab=\"3\" role=\"button\" aria-controls=\"elementor-tab-content-2403\" aria-expanded=\"false\">\n\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon elementor-toggle-icon-left\" aria-hidden=\"true\">\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-closed\"><svg class=\"e-font-icon-svg e-fas-caret-right\" viewBox=\"0 0 192 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M0 384.662V127.338c0-17.818 21.543-26.741 34.142-14.142l128.662 128.662c7.81 7.81 7.81 20.474 0 28.284L34.142 398.804C21.543 411.404 0 402.48 0 384.662z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t<span class=\"elementor-toggle-icon-opened\"><svg class=\"elementor-toggle-icon-opened e-font-icon-svg e-fas-caret-up\" viewBox=\"0 0 320 512\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\"><path d=\"M288.662 352H31.338c-17.818 0-26.741-21.543-14.142-34.142l128.662-128.662c7.81-7.81 20.474-7.81 28.284 0l128.662 128.662c12.6 12.599 3.676 34.142-14.142 34.142z\"><\/path><\/svg><\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t\t<\/span>\n\t\t\t\t\t\t\t\t\t\t\t\t<a class=\"elementor-toggle-title\" tabindex=\"0\">Le\u00e7ons retenues<\/a>\n\t\t\t\t\t<\/div>\n\n\t\t\t\t\t<div id=\"elementor-tab-content-2403\" class=\"elementor-tab-content elementor-clearfix\" data-tab=\"3\" role=\"region\" aria-labelledby=\"elementor-tab-title-2403\"><p><span style=\"font-weight: 400; color: #000000;\">\u201cLa premi\u00e8re le\u00e7on retenue sur un projet data tel que PyKana est que <\/span><span style=\"color: #000000;\"><b>le temps de traitement\/extraction<\/b><\/span> <span style=\"color: #000000;\"><b>des donn\u00e9es de base pour les rendre exploitables est non n\u00e9gligeable.<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> Il est important de s\u2019en rappeler sur tout type de projet, car il faut allouer un cr\u00e9neau important \u00e0 cette \u00e9tape dans le calendrier de celui-ci.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Ensuite, <\/span><span style=\"color: #000000;\"><b>les intuitions que l\u2019on peut avoir sur des r\u00e9sultats ne s\u2019av\u00e8rent pas toujours justes.<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> Nous en avons fait l\u2019exp\u00e9rience lorsque nous avons suppos\u00e9 que nos mod\u00e8les donneraient de meilleurs r\u00e9sultats sur notre base de donn\u00e9es filtr\u00e9es que sur notre base brute. La r\u00e9alit\u00e9 nous a prouv\u00e9 le contraire alors que notre base brute \u00e9tait fortement bruit\u00e9e, cependant peut-\u00eatre que le choix du filtre (Canny) que nous avons fait n\u2019\u00e9tait pas le bon.<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Apr\u00e8s, nous nous sommes vite rendus compte qu\u2019<\/span><span style=\"color: #000000;\"><b>un projet data n\u00e9cessite une r\u00e9partition intelligente des t\u00e2ches au sein de l\u2019\u00e9quipe pour une efficacit\u00e9 accrue.<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> Par exemple, le choix de nous r\u00e9partir s\u00e9par\u00e9ment l\u2019impl\u00e9mentation des diff\u00e9rents mod\u00e8les de Machine\/Deep Learning a \u00e9t\u00e9 d\u00e9terminant pour parall\u00e9liser notre travail et ne pas \u00eatre redondants. Nous avons \u00e9galement op\u00e9r\u00e9 notre r\u00e9partition des t\u00e2ches de fa\u00e7on \u00e0 coller au mieux avec les app\u00e9tences de chacun(e).<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Enfin,<\/span><span style=\"color: #000000;\"><b> l\u2019utilisation d\u2019outils de travail collaboratif est indispensable surtout lorsque l\u2019on travaille de fa\u00e7on parall\u00e8le.<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> Elle permet de versionner son travail et d\u2019\u00e9viter la perte de donn\u00e9es lors de la fusion de codes d\u00e9velopp\u00e9s par deux membres d\u2019une m\u00eame \u00e9quipe.\u201d<\/span><\/p><\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7ead022 elementor-widget elementor-widget-text-editor\" data-id=\"7ead022\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><span style=\"font-weight: 400; color: #000000;\">Merci au groupe Pykana pour son t\u00e9moignage !\u00a0<\/span><\/p><p><span style=\"font-weight: 400; color: #000000;\">Vous aussi vous souhaitez <\/span><span style=\"color: #000000;\"><b>d\u00e9velopper un projet de data science de A \u00e0 Z et \u00eatre encadr\u00e9 par des formateurs experts\u202f?<\/b><\/span><span style=\"font-weight: 400; color: #000000;\"> N\u2019h\u00e9sitez pas \u00e0 prendre rendez-vous en ligne avec notre \u00e9quipe pour discuter de vos objectifs de formation.\u00a0<\/span><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-8e19991 elementor-align-center elementor-widget elementor-widget-button\" data-id=\"8e19991\" data-element_type=\"widget\" data-widget_type=\"button.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"elementor-button-wrapper\">\n\t\t\t<a class=\"elementor-button elementor-button-link elementor-size-sm elementor-animation-grow\" href=\"https:\/\/datascientest.com\/rdv\">\n\t\t\t\t\t\t<span class=\"elementor-button-content-wrapper\">\n\t\t\t\t\t\t<span class=\"elementor-button-text\">Prendre RDV<\/span>\n\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021. Leur objectif est d\u2019utiliser diff\u00e9rents algorithmes de Machine et Deep Learning afin de r\u00e9aliser de la reconnaissance de Kanas, des caract\u00e8res japonais. Comment ont-ils eu l\u2019id\u00e9e du projet\u202f? Ils [&hellip;]<\/p>\n","protected":false},"author":77,"featured_media":177908,"comment_status":"open","ping_status":"open","sticky":false,"template":"elementor_theme","format":"standard","meta":{"content-type":"","inline_featured_image":false},"categories":[2497],"tags":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v21.7 (Yoast SEO v21.7) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)<\/title>\n<meta name=\"description\" content=\"Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)\" \/>\n<meta property=\"og:url\" content=\"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais\" \/>\n<meta property=\"og:site_name\" content=\"Formation Data Science | DataScientest.com\" \/>\n<meta property=\"article:published_time\" content=\"2021-07-22T07:59:20+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-11-09T14:06:52+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/PyKana.png\" \/>\n\t<meta property=\"og:image:width\" content=\"900\" \/>\n\t<meta property=\"og:image:height\" content=\"450\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Raphael Kassel\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"Raphael Kassel\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"13 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais\",\"url\":\"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais\",\"name\":\"PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)\",\"isPartOf\":{\"@id\":\"https:\/\/datascientest.com\/#website\"},\"datePublished\":\"2021-07-22T07:59:20+00:00\",\"dateModified\":\"2023-11-09T14:06:52+00:00\",\"author\":{\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1\"},\"description\":\"Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021.\",\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais\"]}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/datascientest.com\/#website\",\"url\":\"https:\/\/datascientest.com\/\",\"name\":\"Formation Data Science | DataScientest.com\",\"description\":\"Leader de la formation en Data Science\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/datascientest.com\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1\",\"name\":\"Raphael Kassel\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/datascientest.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg\",\"contentUrl\":\"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg\",\"caption\":\"Raphael Kassel\"},\"description\":\"Raphael Kassel est le Head of Data Programs chez DataScientest. Il supervise une \u00e9quipe Data, impl\u00e9mente de nouveaux parcours comme \\\"Chef de projet IA\\\" et \\\"Data Product Manager\\\", et contribue \u00e0 la modernisation des op\u00e9rations du groupe. En collaboration \u00e9troite avec l'\u00e9quipe de d\u00e9veloppement, il adapte l'accompagnement acad\u00e9mique aux \u00e9volutions de l'\u00e9quipe data et \u00e0 l'internationalisation.\",\"sameAs\":[\"https:\/\/www.linkedin.com\/in\/raphaelkassel\/\",\"Data Science\"],\"url\":\"https:\/\/datascientest.com\/author\/raphaelkassel\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)","description":"Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","og_locale":"fr_FR","og_type":"article","og_title":"PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)","og_url":"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","og_site_name":"Formation Data Science | DataScientest.com","article_published_time":"2021-07-22T07:59:20+00:00","article_modified_time":"2023-11-09T14:06:52+00:00","og_image":[{"width":900,"height":450,"url":"https:\/\/datascientest.com\/wp-content\/uploads\/2021\/07\/PyKana.png","type":"image\/png"}],"author":"Raphael Kassel","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"Raphael Kassel","Dur\u00e9e de lecture estim\u00e9e":"13 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","url":"https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais","name":"PyKana - L\u2019application de reconnaissance de caract\u00e8res japonais)","isPartOf":{"@id":"https:\/\/datascientest.com\/#website"},"datePublished":"2021-07-22T07:59:20+00:00","dateModified":"2023-11-09T14:06:52+00:00","author":{"@id":"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1"},"description":"Le projet PyKana est la derni\u00e8re partie de validation de quatre de nos apprenants, Raouf, Franck, Diahana et Marius du bootcamp Data Scientist de Mars 2021.","inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/datascientest.com\/pykana-lapplication-de-reconnaissance-de-kana-caracteres-japonais"]}]},{"@type":"WebSite","@id":"https:\/\/datascientest.com\/#website","url":"https:\/\/datascientest.com\/","name":"Formation Data Science | DataScientest.com","description":"Leader de la formation en Data Science","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/datascientest.com\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/datascientest.com\/#\/schema\/person\/6eca16f63daaafcbc96e15549b9d4ee1","name":"Raphael Kassel","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/datascientest.com\/#\/schema\/person\/image\/","url":"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg","contentUrl":"https:\/\/datascientest.com\/wp-content\/uploads\/2023\/11\/raph.jpeg","caption":"Raphael Kassel"},"description":"Raphael Kassel est le Head of Data Programs chez DataScientest. Il supervise une \u00e9quipe Data, impl\u00e9mente de nouveaux parcours comme \"Chef de projet IA\" et \"Data Product Manager\", et contribue \u00e0 la modernisation des op\u00e9rations du groupe. En collaboration \u00e9troite avec l'\u00e9quipe de d\u00e9veloppement, il adapte l'accompagnement acad\u00e9mique aux \u00e9volutions de l'\u00e9quipe data et \u00e0 l'internationalisation.","sameAs":["https:\/\/www.linkedin.com\/in\/raphaelkassel\/","Data Science"],"url":"https:\/\/datascientest.com\/author\/raphaelkassel"}]}},"_links":{"self":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts\/82060"}],"collection":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/users\/77"}],"replies":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/comments?post=82060"}],"version-history":[{"count":0,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/posts\/82060\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/media\/177908"}],"wp:attachment":[{"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/media?parent=82060"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/categories?post=82060"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/datascientest.com\/wp-json\/wp\/v2\/tags?post=82060"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}