Coursework description: A Tiny Data Science Project You carry out a data science project from start to finish. You will decide on a domain/dataset, come up with a domain-relevant analysis strategy,...

Coursework description: A Tiny Data Science Project

You carry out a data science project from start to finish. You will decide on a domain/dataset, come up with a domain-relevant analysis strategy, perform data wrangling, carry out your analysis/modelling as needed, validate your results and communicate your observations along your journey.

You will broadly follow the whole data science process as discussed in the lectures. So don't just predict something. Dodata analysisand identifyfindingsofrelevance to your application domain

Make sure youenjoythis Tiny Data Science Project! The best projects are those where you enjoy the topic and know something about the application domain.

You will need to deliver:

1. Submission 1: Plan & progress update:a project progress update (unassessed but mandatory) onSunday 3rd November, on which you will get individual feedback.
2. Submission 2: The final analysis: a final submission (computational notebook & report) onSunday 15 December. Yourwhole module markwill be based on this.

Approach

We suggest the following approach:

• Stage 1: Identify the application domain and datasets.You are free to choose the application domain and the dataset that you want to investigate. Some ideas are listed below. Make sure that data are sufficiently complex. Combining one or more datasets may help this so that you can investigate the demonstrate observations that are not obvious by performing trivial operations in a system such as Excel. Source your data from wherever you want (as long as its legal and ethical!). Make sure you state the source of your data. Here are some ideas:

  • A challenging option isYelp Data Challengeprovides datasets on business reviews, social network activities, check-ins, etc. This is an ongoing challenge with a deadline on the last day of the year.(If you're interested in actually entering, please get in touch.)
  • For a financial focus, here is agood resource on industries and companies.You can enrich your analysis by adding data and new perspective from data repositories ofWorld BankandIMF.
  • data.gov.ukis a great place to grab datasets. There's a range in quality, where some is too simple (highly aggregated tables) and some is rather unstructured so might take time to prepare. But generally the data sets are relevant and updated.
  • For a more local focus,London Datastoreis a great place to start. You might even link some of these datasets to the Yelp ones.
  • Airline Data Project is an interesting initiative by MIT where they gathered several sources of information related to the aviation industry. The datasets can beaccessed here. And there is even a glossary to get youintroduced to the domain.
  • Visualizing.org has a nicecollection of data resourceswhich might help you to find relevant domains and datasets.
  • Kaggle competitionsare a great resource to find relevant data sources and problem descriptions. However, if sample solutions for a competition is released, you need to clearly cite any ideas you borrow from the solutions and justify how your solution differs from theirs.

• Stage 2: Identify questions and analysis tasks.Ask yourself"What do I want to achieve/demonstrate in this analysis?".Don't just predict something- you goal is to dodata analysis and identify findingsofrelevance to your application domain. You will be expected to answer this by documenting: a briefoverview of the domain,analytical questionsthat are being asked, the list of yourobjectives, and expected output of your analysis. Familiarise yourself with the problem domain by reading related academic papers, related articles and maybe even interacting with people active in the domain of interest. As the project is "tiny" we don't expect you to be exhaustive, but it will help inform and justify your analysis strategy and findings.

• Stage 3: (Initially explore) and develop an analysis strategy (plan). Your plan should be designed to address the tasks you identify and designed to fulfil your stated objectives in your selected problem domain. This should be informed by aninitial investigationof the data sources and the characteristics of the data. Plan which data processing steps you will need to perform, how you will transform the data to make it useable, which data analysis algorithms you could and what sorts of observations these may lead to. Remember that the data analysis process is highly agile, you will find yourself iterating through several different methods and changing your initial plans frequently.

• Stage 4: Performing the Analysis. This is thecoreof your analysis which will include getting the data ready for analysis, carrying out your analysis/modelling as needed, validating your results and communicating observations. It will involve many iterative steps. Weexpectto see evidence of:

  • data processing(wrangling/merging) to the extent necessary (not all datasets are messy) to prepare useful and robust data to work with
  • data derivationto support your analysis. This may include feature engineering. Be creative
  • building modelsthat help explain the underlying phenomena. These will help you to explain the data better and/or perform predictions. Models include those using regression, classification or clustering methods
  • evaluating (and validate) your resultsusing formal methods and try to minimize uncertainties in your findings.

  Suggestions:

  • Feel free to collect and link additional data, if appropriate.
  • You might consider applying dimension-reduction where you have several data variables and need to find out factors that affect the phenomenon that are not directly visible from raw data.
  • You might consider converting data into a graph and performing some graph theoretical analysis. For instance, transport connections can be treated as a graph and some analysis can be carried out on top.
  • Apply a number of different methods to solve the same analytical tasks and compare results to gain trust in the results.

• Stage 5: Findings and reflections. Identify the findings that are worthy of reflections and reflect on them, articulating them and relating them to the analysis tasks, questions and objectives you have identified. Focus on the most significant ones. How are your observations useful and actionable? Can your observations provide new insight, inform decision making, and if so how? Once you are at this stage, you may still find yourself going back to the previous stages.

General points

• Individual work.This is anindividual piece of work. You work must be carried out individually, though of course you can discuss general concepts about your work with each other and with the teaching team).

• Python and Python notebook. We expect you to present you analysis in aPython notebook. However, feel free to utilize supplementary tools, computational methods, or software to enrich your analysis (include details and files where appropriate in your notebook submission ZIP).

• Analysis not yielding good findings.Not all analysis leads to the observations and findings you expect or wish for. If this happens,don't panic!. As long as the methods and plan you have following is reasonable and you can offer explanations as to why the analysis did not give the results you expected, you will not lose marks. Reasons may include issues to do with data, analysis methods and false expectations. Also, showinga lack of relationshipis also a finding!

• Reference your sources!Clearly reference any resources you make use of in your analysis. This should include URLs of websites that have helped you. This is general good practice. (And the university take this seriously and have a range of sanctions they can apply to people who fluat these.)

• Communicate your findings well.It is in your interest to ensure the marker understands your work and its implications easily.

• Describe the implicationsof your findings to your application domain - i.e. its value.

• Align the stages of your projects.Ensure that the goals, analysis and findings follow from each other. Link your findings to the results of your analytical steps and your objectives.

Submission 1: Plan and progress update (Sunday, 3rd November, 23:55)

Your plan should demonstrate that you've made a start on your project and provide enough detail for useful formative feedback from us. This submission is mandatory, but will not count towards your module marks. It is an opportunity to get individual formative feedback on your idea, plan and its scope. The brief feedback will ensure that your work is progressing in the right direction. It is also designed to encourage you to start work early and already have some initial observations.

Using theJupyter IPython notebooktemplate (see below), you need to demonstrate that you've chosen your data and application domain, have a plan, have started working with the data and have done some initial investigations. As provided by the template, your report needs to have four sections:

• Section 1: Data source, application domain and analytical questions (maximum 150 words):This is a very brief overview ofStages 1 and 2(as described above). Briefly describe your data sources and the domain they are related to. Try to write at leastone key questionyou want answer with your analysis.

• Section 2: Analysis Strategy and Plans (maximum 200 words):This is a very brief overview ofStage 3(described above). Briefly comment on your overall strategy is, what transformations you will need to do, what analytical tools you plan to use.

• Section 3: Initial investigations on the data sources (maximum 150 words):Report some very brief early observations here. Some examples could be: How many features you have, what are the data types, do you observe several problematic data features, will you need significant data transformations? The answers might come from the code you include in the following section.

• Section 4: Python code for initial investigations:Here, we only want you to demonstrate that you've loaded your data into Python for initial investigation and have some ideas of how to approach it. Some code where you load your data and get the first statistical and/or visual summaries to provide you the insight you document in will be sufficient. We just want to see evidence of your progress.

Submission:Download theIPython notebook templateand fill in the required sections. Export the final notebook as an HTML file, zip the whole folder (containing both the HTML and IPython notebook, butnotthe data) and submit as azipfile.

Submission 2: Project Final Submission (Sunday, 15th December, 23:55)

You are expected to deliver your work in two complementary parts:

1. A report: introducing the domain, analytical objectives and your analysis plan, findings and reflections. It needs2 main sections, but feel free to split them into subsections:

   • Analysis Domain, Questions and Plan(maximum 500 words): containing the Stages 1, 2 and 3 as described in the approaches above.
   • Findings and reflections(maximum 1000 words): containing Stages 5 as described in the approaches above. You can include relevant figures from your computational notebook.

   Submission: Submit as a PDF to the submission area

2. A computational notebookThe computational notebook should demonstrate all the steps of your data preparation, analysis and computational modelling. You are expected to present acomputational narrativethat mixes: code, resulting figures, and verbal discussions that explains and comments on the observations made. This should be produced usingJupyter Notebookwritten in Python.

   • Data preparation, analysis and computational modelling(code, figures and a maximum 1500 words of text): this should include Stage 4. Use sub-titles in your notebook to explain the different phases you go through. Make use of visualisations effectively. Justify your analytical steps (e.g. why transform a variable) and discuss observations using clear and well justified arguments.

   Submission:Submit a ZIP of your Python notebook folder containing your .ipynb, an HTML version of it and any support files required.We don't need the data.

Submission 2 will constitute100% of your overall module markand will be based on the computational notebook and the report as a whole. We will look for the following:

• Analysis Domain, Questions, Plan (20%):
  • The text gives aclear and informative overview of the domain
  • The text identifies aclear and convincing motivationfor the analysis
    
  • Analytical problemsare clearly described and listed
  • The report includes awell-considered plan/strategyon how the analysis will be carried out
    
  • Clear objectives for the analysis are presented

• Analytical Process (50%):
  • Awell-considered analysis methodology overallwhich matches well with the motivation
  • Evidence ofeffective data gathering,wranglingandtransformationsteps
  • A set of data transformations and derivation steps adds richness to the analysis
  • A suitable set of analytical tools and methods have been utilised with convincing justifications
  • Incorporateseffective modelling of the datausing one or more of the methods covered, e.g., regression, clustering, dimension red., etc.
  • A number of complementary investigations that build aholistic understanding of the data
• Findings and Reflections (20%):
  • The findingscritically evaluated, discussing any potentialbiasesandlimitations
  • Objectives and overall motivations arerevisitedandcritically evaluated
    
  • Includes reflection on the potential impacts and uses of the findings
  • Evidence of aholistic understandingof the data being analysed
• Clarity and Technical Soundness in Presentation/Argumentation (10%):
  • The report indicatestechnically well-informeddecisions
  • Arguments and analytical steps are properlyjustified
  • Facts and findings are clearly explained and demonstrates agood level of understanding
  • Effective use ofvisuals, charts and factual tables
  • Effective use of references where needed

For postgraduate students, the pass mark is 50%. For undergraduate students, the pass mark is 40%.

General grading criteria