Samenvatting Data Science for Ecology

25% written exam (concept & theory)

25% PC exam (R skills)

50% group work (50% zip, 25% group, 25% ppt)



Concepts & theory
Data science
Top down view: generating value from data



Knowledge pyramid

 Data  info  knowledge  wisdom
 Raw data  meaningful data



Blend of principles & methods

 Ecology (domain) + computer science + maths and statistics



Trends in ecological research

 Large, complex datasets
 Specialised tech
 Data driven multidisciplinary science
 Analysing patterns



OSEMN pipeline

 Obtaining data
 Scrubbing (cleaning) data
 Exploring data
 Modelling data
 INterpreting results



Effective workflow

 Clear data structure
 Concise
 Understandable
 Reproducible
 Transferable

↓

, 1. Import
2. Tidy
3. Transform
4. Visualise
5. Model (transform & visualise again when needed)
6. Communicate



Data science (DS) vs empirical science

 DS based on scientific method
 But: not all data science = science
 Different scale
 Empirical science => small #correlations  causal?
 DS => can identify unlimited #correlations



Data driven vs hypothesis driven

 Data driven
- Inductive
- Starts with data analysis
 Hypothesis driven
- Deductive
- Starts with hypothesis



3Vs of data

 Volume
 Variety
 Velocity

But: DS project can also be based on smaller, simpler data



DS workflow
1. (acquire data)
2. Import
3. Tidy
4. Transform
5. Visualise
6. Model (transform & visualise again when needed)  already 10 steps in
itself
7. Communicate
8. (act)



Gaining insight: transform, visualise, model

, Not 1 template workflow, but: similar steps



Data preparation (1)

 Tidy data in workable format
- Table with rows & columns
- Numeric data
 Convert categorical data  dummy vars
- n classes  n-1 dummy vars
 Deal with missing data
- Remove obs (r)
- Remove vars (c)
- Data imputation
 Correct errors or noise



Feature engineering (2)

 Use domain knowledge to extract features from raw data
 Compute interpretable features/vars from tidy data
- Data mining
 Creativity
- Many features  interactions
- Logical features  simpler models
 Takes lot of time



Algorithm selection (3)

 Problem def
- Classification vs regression  prediction
- Supervised vs unsupervised
- Prediction vs interference
 Multiple algorithms per category/problem
 All algorithms optimise cost/loss function



Feature standardisation (4)

 Centering & scaling
- Standardisation = (x – mean)/sd
- Normalisation = (x – xmin)/(xmax – xmin)
 Improves fit of algorithm
 Improves inference of results



Set division (5)

 Many data points: risk of overfitting