What defines data science?

A multidisciplinary field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from structured and unstructured data.

A technique solely used for data storage and management.

A branch of mathematics that deals with numbers and equations.

A method of programming focused only on software development.

What role does machine learning play in data science?

Machine learning is used to develop algorithms that allow computers to learn from and make predictions based on data.

Machine learning is a tool for graphic design.

Machine learning focuses on manual data entry tasks.

Machine learning is primarily used for data storage.

Which of the following best describes the purpose of data visualization in data science?

To represent data graphically to identify patterns, trends, and insights.

To only create reports for marketing.

To format data for spreadsheet applications.

What is the significance of data cleaning in the data science process?

Data cleaning ensures the accuracy and quality of data by removing or correcting errors and inconsistencies.

Data cleaning simply involves formatting data for better readability.

Data cleaning is only necessary for databases.

Data cleaning does not influence data analysis outcomes.

What is the primary goal of predictive modeling in data science?

To forecast future outcomes based on historical data.

To gather data from the internet.

To create visualizations without statistical tools.

To analyze past data only for reporting purposes.

Which statement best describes big data in the context of data science?

Big data refers to extremely large datasets that require advanced computational methods and tools for processing and analysis.

Big data is only related to small scale data management.

Big data refers to data that is easily processed by standard software.

Big data is a term for data stored in physical files.

What is the role of statistical inference in data science?

To make conclusions about populations based on sample data.

To store data in cloud services.

To collect data from various sources.

What is meant by the term 'data-driven decision making' in data science?

Making decisions based on data analysis and interpretation rather than intuition or personal experience.

Deciding based on past experiences without data consideration.

Making decisions based solely on team discussions.

Relying on opinions from informal polls.

Which technique is commonly used in data science for feature selection?

Recursive Feature Elimination (RFE)

Data Science: Definition

Dès le début de vos recherches, vous allez collecter, produire et exploiter des données. La gestion des données (Research Data Management - RDM) fait partie du processus de recherche. Elle concerne l'ensemble des opérations de collecte, description, stockage, traitement, analyse, archivage et mise en accès des données. (extrait de : Passeport pour la Science Ouverte. Guide pratique pour les doctorants ) "La science ouverte est la diffusion sans entrave des publications et des données de la recherche. Elle s’appuie sur l’opportunité que représente la mutation numérique pour développer l’accès ouvert aux publications et – autant que possible – aux données de la recherche. "Les données de la recherche sont la matière première de la connaissance. Les partager, c'est ouvrir de nouvelles perspectives scientifiques" Source : Plan national pour la Science ouverte - Ministère ESR - Juillet 2018 Source image : https://bibliotheques.univ-tlse3.fr/file/composantes-science-ouverte Cette page est une introduction à la gestion des données de recherche. Elle présente quelques concepts et étapes clés pour vous engager dans cette démarche. Consultez les liens pour approfondir vos connaissances. • What are data ? Définition des données de recherche de l’OCDE (2007) « Enregistrements factuels (chiffres, textes, images, sons) utilisés comme source principale pour la recherche scientifique et généralement reconnus par la communauté scientifique comme nécessaires pour valider les résultats de la recherche. Un ensemble de données de recherche constitue une représentation systématique et partielle du sujet faisant l’objet de la recherche ». Exemples • les images d’une ville préhistorique deviennent des données pour un chercheur qui étudie l’histoire de cette ville; • les « données » d’un linguiste peuvent être des écrits ou des discours, des enregistrements de locuteurs ; • les « données » d’un médiéviste sont des sources archivistiques, archéologiques, épigraphiques, iconographiques, littéraires ; • les « données » d’un géologue rassemblent des coupes et observations de terrain consignées sur un carnet, des résultats de carottage, des analyses d’échantillons, des données sismographiques… • • Pourquoi partager ses données ? "La science ouverte vise à construire un écosystème dans lequel la science est plus cumulative, plus fortement étayée par des données, plus transparente, plus rapide et d’accès plus universel.La science ouverte favorise également les avancées scientifiques, particulièrement les avancées imprévues, ainsi que l’innovation, les progrès économiques et sociaux, en France, dans les pays développés et dans les pays en développement. Enfin, la science ouverte constitue un levier pour l’intégrité scientifique et favorise la confiance des citoyens dans la science. Elle constitue un progrès scientifique et un progrès de société." Source : Plan national pour la Science Ouverte (2018) Les enjeux de l'Open Data • enjeux patrimoniaux o preuve et mémoire (éviter les pertes de données) • enjeux économiques o valeur économique de la donnée o réutilisation gratuite ou payante des données, exploitation des résultats de recherches antérieures (éviter de refaire ce qui a déjà été validé), o accélération de l'innovation et le retour sur investissement dans la R&D • enjeux scientifiques o de "hypothesis-driven" à "data-driven" o plus de visibilité pour le scientifique • enjeux sociétaux o participation des citoyens et de la société civile : "Citizen science" o confiance en la recherche Pour aller plus loin • Site Doranum : https://doranum.fr/enjeux-benefices/fiche-synthetique/ • Adopter de bonnes pratiques tout au long du cycle de vie des données De bonnes pratiques de gestion à toutes les étapes du cycle de vie de la donnée sont un préalable indispensable à l’ouverture des données et à leur réutilisation. • Rechercher des données Pour identifier des jeux de données (datasets) pertinents pour votre thèse, des outils de recherche sont disponibles. Suivez ces liens pour les découvrir : • Site Doranum : https://doranum.fr/acces-visualisation/rechercher-donnees/ • Site DataCC - Vos besoins, trouver des données : https://www.datacc.org/vos-besoins/trouver-des-donnees/ • Fiche CoopIST : Trouver des jeux de données via des bases pluridisciplinaires et des moteurs de recherche Pensez-aussi à consulter l'entrepôt institutionnel Data INRAE Page de présentation du portail • Choisir les bons formats et bien organiser vos données  Choisir des formats de fichier : https://www6.inrae.fr/datapartage/Gerer/Choisir-des-formats-de-fichier  Nommer et organiser vos fichiers de données : https://www6.inrae.fr/datapartage/Gerer/Nommer-et-organiser-ses-fichiers-de-donnees Pour aller plus loin • Jaouen, G.- Gérer ses données. Pourquoi, Comment ? Séminaire - Guadeloupe, du 25 au 27 Novembre 2019 – CRAG INRA • Bien décrire et documenter ses données La description d’un jeu de données se fait à l’aide de métadonnées (*) qui doivent apporter suffisamment d'éléments (sur la collecte des données, les unités de mesure employées...) pour chercher et trouver le jeu de données, juger de sa qualité/fiabilité, et pouvoir le comprendre ou le réutiliser dans un autre contexte. (*) Définition des métadonnées : Ensemble d’informations structurées qui décrit, explicite, localise une ressource informationnelle, dans le but d’en faciliter la recherche, l’usage, et la gestion. Source : NISO. Understanding Metadata. 2004. Quelques liens utiles : • Site Doranum : https://doranum.fr/metadonnees-standards-formats/ • DataCC : https://www.datacc.org/vos-besoins/documenter-ses-donnees/metadonnees/ • Site DataPartage INRAE : https://www6.inrae.fr/datapartage/Gerer/Documenter-les-donnees En complément des métadonnées, la rédaction d'un fichier READ ME.txt est également recommandée. • Stocker, sécuriser, préserver ses données Bien différencier les notions de stockage et d'archivage. Anticiper pour déterminer les données à éliminer et celles qui doivent être préservées à long terme. • Dans l'environnement INRAE : https://www6.inrae.fr/datapartage/Gerer/Stocker-les-donnees • Site Doranum : https://doranum.fr/stockage-archivage/ • Site DataCC : https://www.datacc.org/vos-besoins/conserver-ses-donnees/ • Partager, ne pas partager ses données ? Dans le cadre de la Science Ouverte, il y a de plus en plus d'incitations voire d'exigences pour rendre accessibles les données, en particulier les données liées aux publications : • de l'édition scientifique : de plus en plus de revues adoptent une "data policy" (à consulter dans les instructions aux auteurs) et exigent des auteurs qu'ils fournissent les données associées aux publications, • des organismes de financement (ANR, Commission Européenne ...), • des politiques nationale (Plan national pour la Science ouverte - Ministère ESR - Juillet 2018) et institutionnelle. Mais attention, toutes les données ne sont pas partageables : assurez-vous que vos données sont bien diffusables au regard du droit et des conditions d'exercice de votre thèse et de son mode de financement (se reporter à votre contrat de thèse). Les données produites dans les organismes de recherche publics sont communicables à tous si elles n'entrent pas dans le cadre d'exceptions légales (sécurité défense, sécurité des populations, patrimoine scientifique et technique, données personnelles, données liées au secret, statistique, etc.) Liens utiles : • sur le site Data Partage, la page Partager-Publier ou la page : "Données de la recherche : qui a les droits, qui doit partager ?" • le site INRAE dédié à la protection des données personnelles et l'application du RGPD (Règlement général sur la protection des données) : https://intranet.inrae.fr/cil-dpo • Valoriser ses données Voici les principales voies de diffusion •  Partager ses données en les déposant dans un entrepôt  Choisir un entrepôt  Déposer dans Data INRAE  Partager ses données comme matériel supplémentaire d'un article (à la demande de l'éditeur)  Publier un Data Paper (article de données) : la meilleure voie en terme de visibilité des données, et pour faciliter leur réutilisation. Pour aller plus loin • Site Doranum o Dépôts et entrepôts. Comment et où déposer mes données ? o Data papers et Data journals. Comment publier mes données comme un article scientifique ? • Site DataCC o Valoriser ses données • Site CoopIST o Déposer des données de recherche dans un entrepôt o Rédiger et publier un data paper dans une revue scientifique A télécharger : Synthèse du processus de rédaction d'un article avec des données associées • Pourquoi ne pas rédiger un plan de gestion de données (PGD) pour votre thèse ? La thèse peut être assimilée à un projet et certaines universités au Royaume Uni, aux Pays-Bas et plus récemment en France préconisent la rédaction d'un plan de gestion associé à la thèse. Le PGD (ou DMP = Data Management Plan) est un outil de planification qui peut vous aider à anticiper et bien gérer toutes les étapes du cycle de vie de vos données, à limiter les risques de perte ou corruption de données, à adopter de bonnes pratiques de gestion, pour in fine produire des données respectueuses des principes FAIR, adoptés aujourd'hui par l'ensemble des acteurs de la recherche. Il est désormais exigé par la plupart des financeurs de la recherche (Commission Européenne et ANR ...) dans le cadre de projets financés. Rédiger un PGD pour votre thèse, peut être un bon exercice pour vous préparer à la future rédaction de réponses à des appels d'offre. Comment faire en pratique ? • Site DataPartage : Pourquoi et comment rédiger un plan de gestion de données ? • Site Doranum : https://doranum.fr/plan-gestion-donnees-dmp/, La minute vidéo PGD • Site DataCC : https://www.datacc.org/bonnes-pratiques/adopter-un-plan-de-gestion-des-donnees/ • Suivre une classe virtuelle INRAE : Open Class "Rédaction d'un PGD" • Produire des données FAIR ! Favoriser la production de données FAIR (Findable - Accessible - Interoperable - Reusable) est aujourd'hui un objectif soutenu par l'ensemble des acteurs de la recherche. Source : https://open-science-training-handbook.gitbook.io/book/ Si vous suivez les conseils et recommandations de cette page, vous avez toutes les chances d'avoir produit des données de qualité. Si vous préférez une version illustrée : "Pensez FAIR" - https://datapartage.inrae.fr/Gerer/Cycle-de-la-donnee Affiche cycle de vie des données réalisée dans le cadre des Missions QualiNous & RGPD, INRAE-ACT Vous pouvez tester le niveau de "Fairification" de vos données grâce à ces outils :  ARDC : https://ardc.edu.au/resources/working-with-data/fair-data/fair-self-assessment-tool • D'autres ressources pour se former ou s'autoformer En interne INRAE • Formation à la science ouverte OSCAR - Module "Gestion et partage des données" • Le site "Gestion et partage des données" • Des classes virtuelles d'environ 2h (Open Class) sont régulièrement proposées sur : o la rédaction des plans de gestion de données, o le dépôt et la description d'un jeu de données dans Data INRAE, o la rédaction et la publication de data papers, Sites externes • Le site DORANUM (Données de la Recherche : Apprentissage NUMérique à la gestion et au partage) propose un dispositif de formation à distance intégrant de nombreuses ressources d’auto-formation déclinées sur différents supports (textes, infographies, vidéos) et sur 9 thématiques. o Parcours interactif sur la gestion des données de la recherche (2020) o • Le site DataCC. Accompagnement à la gestion des données de recherche en physique et en chimie : https://www.datacc.org/ o Data Stories : https://www.datacc.org/reseau-datacc/data-stories/ o • Le dossier "Open Access & Open Data" réalisé par l'Ecole des Ponts - ParisTech • • The Open Science Training Handbook : https://www.ouvrirlascience.fr/the-open-science-training-handbook/

“There’s No Such Thing as Sound Science” by By Christie Aschwanden was a lead science writer for FiveThirtyEight. FiveThirtyEight, Science, Dec. 6, 2017 Science is being turned against itself. For decades, its twin ideals of transparency and rigor have been weaponized by those who disagree with results produced by the scientific method. Under the Trump administration, that fight has ramped up again. In a move ostensibly meant to reduce conflicts of interest, Environmental Protection Agency Administrator Scott Pruitt has removed a number of scientists from advisory panels and replaced some of them with representatives from industries that the agency regulates. Like many in the Trump administration, Pruitt has also cast doubt on the reliability of climate science. For instance, in an interview with CNBC, Pruitt said that “measuring with precision human activity on the climate is something very challenging to do.” Similarly, Trump’s pick to head NASA, an agency that oversees a large portion the nation’s climate research, has insisted that research into human influence on climate lacks certainty, and he falsely claimed that “global temperatures stopped rising 10 years ago.” Kathleen Hartnett White, Trump’s nominee to head the White House Council on Environmental Quality, said in a Senate hearing last month that she thinks we “need to have more precise explanations of the human role and the natural role” in climate change. The same entreaties crop up again and again: We need to root out conflicts. We need more precise evidence. What makes these arguments so powerful is that they sound quite similar to the points raised by proponents of a very different call for change that’s coming from within science. This other movement strives to produce more robust, reproducible findings. Despite having dissimilar goals, the two forces espouse principles that look surprisingly alike: Science needs to be transparent. Results and methods should be openly shared so that outside researchers can independently reproduce and validate them. The methods used to collect and analyze data should be rigorous and clear, and conclusions must be supported by evidence. These are the arguments underlying an “open science” reform movement that was created, in part, as a response to a “reproducibility crisis” that has struck some fields of science.1 But they’re also used as talking points by politicians who are working to make it more difficult for the EPA and other federal agencies to use science in their regulatory decision-making, under the guise of basing policy on “sound science.” Science’s virtues are being wielded against it. What distinguishes the two calls for transparency is intent: Whereas the “open science” movement aims to make science more reliable, reproducible and robust, proponents of “sound science” have historically worked to amplify uncertainty, create doubt and undermine scientific discoveries that threaten their interests. “Our criticisms are founded in a confidence in science,” said Steven Goodman, co-director of the Meta-Research Innovation Center at Stanford and a proponent of open science. “That’s a fundamental difference — we’re critiquing science to make it better. Others are critiquing it to devalue the approach itself.” Calls to base public policy on “sound science” seem unassailable if you don’t know the term’s history. The phrase was adopted by the tobacco industry in the 1990s to counteract mounting evidence linking secondhand smoke to cancer. A 1992 Environmental Protection Agency report identified secondhand smoke as a human carcinogen, and Philip Morris responded by launching an initiative to promote what it called “sound science.” In an internal memo, Philip Morris vice president of corporate affairs Ellen Merlo wrote that the program was designed to “discredit the EPA report,” “prevent states and cities, as well as businesses from passing smoking bans” and “proactively” pass legislation to help their cause. The sound science tactic exploits a fundamental feature of the scientific process: Science does not produce absolute certainty. Contrary to how it’s sometimes represented to the public, science is not a magic wand that turns everything it touches to truth. Instead, it’s a process of uncertainty reduction, much like a game of 20 Questions. Any given study can rarely answer more than one question at a time, and each study usually raises a bunch of new questions in the process of answering old ones. “Science is a process rather than an answer,” said psychologist Alison Ledgerwood of the University of California, Davis. Every answer is provisional and subject to change in the face of new evidence. It’s not entirely correct to say that “this study proves this fact,” Ledgerwood said. “We should be talking instead about how science increases or decreases our confidence in something.” The tobacco industry’s brilliant tactic was to turn this baked-in uncertainty against the scientific enterprise itself. While insisting that they merely wanted to ensure that public policy was based on sound science, tobacco companies defined the term in a way that ensured that no science could ever be sound enough. The only sound science was certain science, which is an impossible standard to achieve. “Doubt is our product,” wrote one employee of the Brown & Williamson tobacco company in a 1969 internal memo. The note went on to say that doubt “is the best means of competing with the ‘body of fact’” and “establishing a controversy.” These strategies for undermining inconvenient science were so effective that they’ve served as a sort of playbook for industry interests ever since, said Stanford University science historian Robert Proctor. The sound science push is no longer just Philip Morris sowing doubt about the links between cigarettes and cancer. It’s also a 1998 action plan by the American Petroleum Institute, Chevron and Exxon Mobil to “install uncertainty” about the link between greenhouse gas emissions and climate change. It’s industry-funded groups’ late-1990s effort to question the science the EPA was using to set fine-particle-pollution air-quality standards that the industry didn’t want. And then there was the more recent effort by Dow Chemical to insist on more scientific certainty before banning a pesticide that the EPA’s scientists had deemed risky to children. Now comes a move by the Trump administration’s EPA to repeal a 2015 rule on wetlands protection by disregarding particular studies. (To name just a few examples.) Doubt merchants aren’t pushing for knowledge, they’re practicing what Proctor has dubbed “agnogenesis” — the intentional manufacture of ignorance. This ignorance isn’t simply the absence of knowing something; it’s a lack of comprehension deliberately created by agents who don’t want you to know, Proctor said.2 In the hands of doubt-makers, transparency becomes a rhetorical move. “It’s really difficult as a scientist or policy maker to make a stand against transparency and openness, because well, who would be against it?” said Karen Levy, researcher on information science at Cornell University. But at the same time, “you can couch everything in the language of transparency and it becomes a powerful weapon.” For instance, when the EPA was preparing to set new limits on particulate pollution in the 1990s, industry groups pushed back against the research and demanded access to primary data (including records that researchers had promised participants would remain confidential) and a reanalysis of the evidence. Their calls succeeded and a new analysis was performed. The reanalysis essentially confirmed the original conclusions, but the process of conducting it delayed the implementation of regulations and cost researchers time and money. Delay is a time-tested strategy. “Gridlock is the greatest friend a global warming skeptic has,” said Marc Morano, a prominent critic of global warming research and the executive director of ClimateDepot.com, in the documentary “Merchants of Doubt” (based on the book by the same name). Morano’s site is a project of the Committee for a Constructive Tomorrow, which has received funding from the oil and gas industry. “We’re the negative force. We’re just trying to stop stuff.” Some of these ploys are getting a fresh boost from Congress. The Data Quality Act (also known as the Information Quality Act) was reportedly written by an industry lobbyist and quietly passed as part of an appropriations bill in 2000. The rule mandates that federal agencies ensure the “quality, objectivity, utility, and integrity of information” that they disseminate, though it does little to define what these terms mean. The law also provides a mechanism for citizens and groups to challenge information that they deem inaccurate, including science that they disagree with. “It was passed in this very quiet way with no explicit debate about it — that should tell you a lot about the real goals,” Levy said. But what’s most telling about the Data Quality Act is how it’s been used, Levy said. A 2004 Washington Post analysis found that in the 20 months following its implementation, the act was repeatedly used by industry groups to push back against proposed regulations and bog down the decision-making process. Instead of deploying transparency as a fundamental principle that applies to all science, these interests have used transparency as a weapon to attack very particular findings that they would like to eradicate. Now Congress is considering another way to legislate how science is used. The Honest Act, a bill sponsored by Rep. Lamar Smith of Texas,3 is another example of what Levy calls a “Trojan horse” law that uses the language of transparency as a cover to achieve other political goals. Smith’s legislation would severely limit the kind of evidence the EPA could use for decision-making. Only studies whose raw data and computer codes were publicly available would be allowed for consideration. That might sound perfectly reasonable, and in many cases it is, Goodman said. But sometimes there are good reasons why researchers can’t conform to these rules, like when the data contains confidential or sensitive medical information.4 Critics, which include more than a dozen scientific organizations, argue that, in practice, the rules would prevent many studies from being considered in EPA reviews.5 It might seem like an easy task to sort good science from bad, but in reality it’s not so simple. “There’s a misplaced idea that we can definitively distinguish the good from the not-good science, but it’s all a matter of degree,” said Brian Nosek, executive director of the Center for Open Science. “There is no perfect study.” Requiring regulators to wait until they have (nonexistent) perfect evidence is essentially “a way of saying, ‘We don’t want to use evidence for our decision-making,’” Nosek said. Most scientific controversies aren’t about science at all, and once the sides are drawn, more data is unlikely to bring opponents into agreement. Michael Carolan, who researches the sociology of technology and scientific knowledge at Colorado State University, wrote in a 2008 paper about why objective knowledge is not enough to resolve environmental controversies. “While these controversies may appear on the surface to rest on disputed questions of fact, beneath often reside differing positions of value; values that can give shape to differing understandings of what ‘the facts’ are.” What’s needed in these cases isn’t more or better science, but mechanisms to bring those hidden values to the forefront of the discussion so that they can be debated transparently. “As long as we continue down this unabashedly naive road about what science is, and what it is capable of doing, we will continue to fail to reach any sort of meaningful consensus on these matters,” Carolan writes. The dispute over tobacco was never about the science of cigarettes’ link to cancer. It was about whether companies have the right to sell dangerous products and, if so, what obligations they have to the consumers who purchased them. Similarly, the debate over climate change isn’t about whether our planet is heating, but about how much responsibility each country and person bears for stopping it. While researching her book “Merchants of Doubt,” science historian Naomi Oreskes found that some of the same people who were defending the tobacco industry as scientific experts were also receiving industry money to deny the role of human activity in global warming. What these issues had in common, she realized, was that they all involved the need for government action. “None of this is about the science. All of this is a political debate about the role of government,” she said in the documentary. These controversies are really about values, not scientific facts, and acknowledging that would allow us to have more truthful and productive debates. What would that look like in practice? Instead of cherry-picking evidence to support a particular view (and insisting that the science points to a desired action), the various sides could lay out the values they are using to assess the evidence. For instance, in Europe, many decisions are guided by the precautionary principle — a system that values caution in the face of uncertainty and says that when the risks are unclear, it should be up to industries to show that their products and processes are not harmful, rather than requiring the government to prove that they are harmful before they can be regulated. By contrast, U.S. agencies tend to wait for strong evidence of harm before issuing regulations. Both approaches have critics, but the difference between them comes down to priorities: Is it better to exercise caution at the risk of burdening companies and perhaps the economy, or is it more important to avoid potential economic downsides even if it means that sometimes a harmful product or industrial process goes unregulated? In other words, under what circumstances do we agree to act on a risk? How certain do we need to be that the risk is real, and how many people would need to be at risk, and how costly is it to reduce that risk? Those are moral questions, not scientific ones, and openly discussing and identifying these kinds of judgment calls would lead to a more honest debate. Science matters, and we need to do it as rigorously as possible. But science can’t tell us how risky is too risky to allow products like cigarettes or potentially harmful pesticides to be sold — those are value judgements that only humans can make.

Data Science

Data Science Data Insights

Data Science Structured vs Unstructed Data Sets

Data Science Privacy Issues

Related quizzes

Related quizzes