Data science life cycle final
- 1. Manoj Mishra November 23, 2017 DataScience Lifecycle
- 2. Introduction LifeCycle SkillTree Questions AGENDA
- 3. DATA SCIENTIST 60% 19% 9% 7% 5% Effort Organize & Clean Data Collect data / Dataset Data Mining to draw pattern Model Selection , training and refining Other Tasks
- 4. Data Acquisition Data Preparation Hypothesis & Modelling DATA SCIENCE LIFE CYCLE Evaluation & Interpretation DeploymentOperations Optimization Business Understanding
- 5. DATA ACQUISITION Static • Feedback system • CSV Data sets / text files Live • Logs data, memory dumps • Sensors, controllers etc. Virtual • Data Virtualization • Caching , Storing
- 6. DATA SAMPLE DATASET INVENTORY
- 7. PROJECT - PREDICTING FAILURE – PROACTIVE MAINTENANCE • Baseline normal operational patterns by modelling the unstructured Log data. • Use Domain Experts to identify patterns before failures. • Use statistical measurements & Machine Learning to determine threshold. • Identify patterns of activity to anticipate and react to circumstances that might otherwise disrupt operations SME - Domain
- 8. DATA PREPARATION • Need for Data Preparation • Bad data or poor quality data can alter accuracy & lead to incorrect Insights • Gartner- Poor quality data costs an avg. organization $13.5M / year. • Dataset might contain discrepancies in the names or codes. • Dataset might contain outliers or errors. • Dataset lacks your attributes of interest for analysis. • All in all the dataset is not qualitative but is just quantitative. • Steps Involved
- 9. DATA PREPARATION • Includes steps to explore, preprocess, and condition data • Create robust environment – analytics sandbox • Data preparation tends to be the most labor-intensive step in the analytics lifecycle • Often at least 50 – 60% of the data science project’s time • The data preparation phase is generally the most iterative and the one that data scientists tend to underestimate most often
- 10. Database : Understand the data Understand the Business Airlines : NYC FLIGHTS 13 e.g. tailnum : A tail number refers to an identification number painte d on an aircraft, frequently on the tail Goal : Predicting flight delays Modelling
- 11. PREDICTING FLIGHT DELAYS - NYC FLIGHTS 13 • Exploratory Data Analysis of the flight data for inbound and outbound flights for year 2013 in NYC. • Find patterns, benchmark, model and find predictors. • To predict the flight delays for NYC Inbound/ Outbound flights. • Ref. DataSet
- 12. PREDICTING FLIGHT DELAYS - NYC FLIGHTS 13 • Exploratory Data Analysis of the flight data for inbound and outbound flights for year 2013 in NYC. • Find patterns, benchmark, model and find predictors. • To predict the flight delays for NYC Inbound/ Outbound flights. • Ref. DataSet
- 13. OBSERVATIONS - NYC FLIGHTS 13 REF. DATASET
- 14. COMMON TOOLS - FOR DATA PREPARATION • Alpine Miner provides a graphical user interface for creating analytic workflows • OpenRefine (formerly Google Refine) is a free, open source tool for working with messy data • Similar to OpenRefine, Data Wrangler is an interactive tool for data cleansing an transformation • Alteryx and Informatica also can be tried.
- 15. HYPOTHESIS & MODELLING • There are three main tasks addressed in this stage: • Feature engineering: Create data features from the raw data to facilitate model training. • Model training: Find the model that answers the question most accurately by comparing their success metrics. • Determine if your model is suitable for production.
- 16. FEATURE SELECTION & ENGINEERING Select Features Research feature relevance Experiment & Validate Change the feature set if required Go back to feature selection
- 17. FEATURE ENGINEERING Date # footfalls in Dubai Mall 01/07/2017 124532 02/07/2017 65434 03/07/2017 12333 04/07/2017 60009 05/07/2017 46567 06/07/2017 98001 07/07/2017 146543 08/07/2017 112345 09/07/2017 76543 Date # footfalls in Dubai Mall IsHoliday? 01/07/2017 124532Yes 02/07/2017 65434No 03/07/2017 12333No 04/07/2017 60009No 05/07/2017 46567No 06/07/2017 98001No 07/07/2017 146543yes 08/07/2017 112345yes 09/07/2017 76543No
- 18. FEATURE ENGINEERING Seasonality (holiday season): Jun, Jul & Dec account for the highest avg. delays
- 19. MODELLING
- 20. CREATE YOUR MODEL & EVALUATE • Split the input data randomly for modeling into a training data set and a test data set. • Build the models by using the training data set. • Evaluate the training and the test data set. Use a series of competing machine- learning algorithms along with the various associated tuning parameters (known as a parameter sweep) that are geared toward answering the question of interest with the current data. • Determine the “best” solution to answer the question by comparing the success metrics between alternative methods.
- 21. CREATE YOUR MODEL & EVALUATE • Supervised Learning • Naive Bayes • KNN • Support Vector Machines (SVM) • Linear Regression • Unsupervised Learning • Principal Component Analysis. • K Means • Classification Metrics • Accuracy Score • Classification Report • Confusion Matrix • Regression Metrics • Mean Absolute Error. • Mean Squared Error • R2 Score • Clustering Metrics • Adjusted Rand Index. • Homogeneity • V - measure
- 22. DEPLOYMENT After you have a set of models that perform well, you can operationalize them for other applications through APIs or other interface to consume from various applications, such as: • Online websites • Spreadsheets • Dashboards • Line-of-business applications • Back-end applications
- 24. THANK YOU !