Guest blog post.

After reading many blog posts, articles and books, I have collected ingredients of data science! Moreover, I've classified them with a purpose of easily making a cook named as data science with below lists for whom wants to construct own career road map! Maybe, you wonder "why I am not giving the recipes of it" is because I do not have any real life experience.        

Creative mix of basic ingredients - not unlike data science - to produce delicious curried black spaghetti 

Firstly, we most probably have encountered a person with pure data warehousing or machine learning background identifying them as data scientist. This is because their areas highly nested with ds. But, what about other ds, mostly, related areas:

Data science related areas           

  • Statistics              
  • Social Science            
  • Database Querying      
  • Data Warehousing            
  • Data Mining        
  • Machine Learning       
  • Bio - Informatics            
  • Business/Data Analyst         

If we look more closely at any of these areas, we'll be more likely to grasp basics of ds, and to meet core activities done on data scientists' daily life. For this purpose, let's look at general tasks done on data mining:

General data mining tasks           

  • Classification        
  • Similarity Model      
  • Profiling                                      
  • Data Reduction        
  • Causal Model            
  • Regression            
  • Clustering                  
  • Co - Occurrence Grouping      
  • Link Prediction          

Until now, I assume I have, implicitly, give some intuition to you, and I'm sure you have already seen a writings about `What is Data Science?`. However, outside the definitions done on any sources, let's take a look at ingredients, skills, of it. While doing this, don't forget the dynamics of life; that is what I have listed below is likely to change over time.  

Skills and technical knowledge required 

  • Hacking Skill / Analytic Thinking      
  • Presentation Skill + Story Telling + Communication Skill            
  • Strategy Expertise                              
  • Mathematics + Statistics + Computer Science Background            
  • Machine Learning' s ,  Statistics', or Mathematics'  Data Modeling Skill            
  • Exploratory Data Analysis: Interactive Data Visualization, Feature Selection & Extraction, EDA Done On Current Data Types  (Traditional e.g. Numerical, Categorical, Binary;  Text  e.g. Email, Tweet, Article;  Record e.g. Timestamped Event Data, JSON / BSON Formatted Files; Geographic Based Location Data; Sensor Data, e.g. RFID / RF Tags Data; Online/Offline Network; Images            
  • Data Mining Algorithm Skill            
  • Performance Visualization & Common Test Methods: Performance Metrics & Tools, Trade Off (e.g. Bias - Variance, Detection Error), Matrix (e.g. Contingency, Cost - Benefit), Evaluation Metrics (e.g. FP Rate, TP Rate, F-Measure, Sensitivity, Recall, Precision, etc), Methods for Estimating Generalization Error (Hold-Out, Cross Validation, Random Sampling, Bootstrap, K-fold / Progressive Cross Validation), Graphs (Lift Curve, Learning  Graph, ROC & AUC, Calibration Graph, Fitting Graph, Loss Comparison Graph, Precision - Recall Graph, PN Graph, Cost Curve),Tests (Density Estimate Test), Causality Tests (Simpson's Paradox, A/B Test, Observing Data Distribution, Robin Causal Model, Causal - Effect Test, Observational Study)            
  • Skill Set From DS Related Areas According To Interest such as Bio - Informatics            


  • Areas (Virtualization, Social Computing e.g. Social Network Analysis, Big Data, Mobile Computing, Parallel & Distributed Computing)                                             
  • Technologies (Hadoop & Map Reduce e.g. HDFS, Hive, Pig, Zookeeper)                                
  • Databases (Relational DB e.g. PostgreSQL, Document DB e.g. MongoDB, Key-Value DB e.g. RIAK, Columnar DB e.g. HBase, Graph DB e.g. Neo4j, Spatial DB e.g. PostGis)                       
  • Computer Programming Skills (R, Octave/Matlab, Java  e.g. Java PL / Json / Bson,  Python, Unix Shell Skills, NoSQL, Social Network Analysis e.g. iGraph, XNetwork, Pejak, NetLogo)        

As an additional part, I believe "how you start and end your activities" will highly correlate with quality of your work. Therefore, before putting hand on real data science job, it seems to be armed with theoretical background on below key, common, points be rescue many life from playing around undeployable applications. Here, I'm willing to light on common pitfalls raised from not considering;            

  • Data Analysis Phase: Base Line Performance - Sparseness - Unconsciously Done Stratification - Curse of Dimensionality - Relative Importance of Features - Measurement Errors -  Correlated Features - Missing Data Values  - Data Leakage - Feature Selection & Extraction (Selection Algorithm  e.g. Forward, Backward, Hybrid; Selection Criteria e.g. R Squared, P Value, AIC,  BIC, Entropy; Sample / Population:                                       Aware Of Ways Of "How To Take Sample", Avoiding Biases In Sample,  Awaring of "How Much Data For Sample Enough To Learn Features")           
  • Model Evoluation Phase: Real-Time Computational Complexity of Scalable Data;  Aware Of "Preferences Change Over Time"; Cost of Model Update; Real-Time Auto Tuning Capability; Enabling To Take More, Better, Faster Machine Learning, Statistical, or Mathematical; Model Experimentation On Data            
  • Model Performance Visualization Phase: Sensitivity snd Limitations of Evaluation Metrics; Bias; Variance Tradeoff / Induction; Deduction Tradeoff;  Determine "What Part of Pipeline Should Be Spent Most Of Time For Pipelined -  Complex System?" e.g Ceiling Analysis, Ablative Analysis        

Even if reaching to end of the post, we still need a set of practices and techniques in order to be gradually close to deployable solution of our planned product. In twitter, data science blogs or associations, they somehow agree on Agile as data scientists' methodology. I think one of the reasons is embedded cycling bahavior of DS. By cycling behavior, I mean we start from basic model, doing experiences, tuning knobs, and then analyzing test results. Afterwards, we continue with decision done on experiences and test results until deployable product. As a last point, here Agile methodologies are;            

  • Scrum                        
  • Unified Process            
  • Agile Model              
  • Disciplined Agile Delivery                

As a final word, I have done this research to clear my career road map and put light on each dark corner. It helps me a lot, and I wish it will help you, too. However, I admit this blog post needs more attention than what I've spend. So, please feel free to comment.          

DSC Resources

Additional Reading

Follow us on Twitter: @DataScienceCtrl | @AnalyticBridge

Views: 2147


You need to be a member of Data Science Central to add comments!

Join Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service