An implementation of a fun weather simulator in Scala programming language.

The fun journey continues.

Technology stack
   - Scala 
   - Apache Spark MLlib 
   - Predictive modelling 
   - Random Forests (Classification and Regression) 
   - LIBSVM data format for training data

Key features
   - Increased model accuracy by increaing training data size and quality 
   - Flexible for future extensions 

Simulation methodology 
   - The environment (atmosphere, topography, geography and oceanography) is summarised by the
     training samples with the attributes of position, time, weather conditions and sensor measurements  
   - Struture the enviroment and observations into training sets (in LIBSVM format):
       - Condition.txt
       - Temperature.txt
       - Pressure.txt
       - Humidity.txt
   - Build classification model for weather conditions 
   - Build regression models for weather sensors measurements (Temperature/Pressure/Humidity) 
   - Feed new data (time&location) the established predictive models to generate the simulated weather data

Usage
   - sbt test or ./activator test 
   - sbt run or ./activator run
   - sbt "run-main com.funweather.WeatherSimulator"
   - sbt "run-main com.funweather.WeatherSimulator [latitude] [longitude] [elevation] [time stamp] 
   - Users may also want to modify or add test scenarios in WeatherSimulator.scala and ./activator run from terminal
   - Open the sbt project using your favorite IDE and run the tests and WeatherSimulator 
        - If IntelliJ encounters error 
                thread "main" java.lang.IllegalArgumentException: System memory 259522560 must be at least 4.718592E8. Please ...
          set the VM options in Run/Debug Configurations to: -Xms128m -Xmx512m -XX:MaxPermSize=300m -ea

TODO
   - Use R package rscala to integrate R scripts in fun-with-weather-R to produce training data and for
     data visualisation
   - Add training set from real weather data to make the simulation more realistic 
   - Feature engineering to extract more information from the time stamps, i.e. hour of the day, month of the 
     year to increase model accuracy
     
Miscellaneous
   - Random Forest
        The pick of the Machine Learning algorithm is Random Forest:
             1. Machine Learning
                > Random Forest can be less prone to overfitting. Training more trees in a Random Forest reduces
                  the likelihood of overfitting because of the variance reduction from avaraging multiple trees in the forest
                > Invariant under scaling and various other transformations of feature values, is robust to inclusion of
                  irrelevant features, and produces inspectable models
                > Random forests can be used to rank the importance of variables in a regression or classification problem
                  in a natural way
             2. Software Engineering
                > One of the most successful machine learning models for classification and regression
                > Ensembles of decision trees; simple numerical optimization; easy to understand and implement
                > Easy tunning for peformance (accuracy and speed)
             3. Big Data
                > Can be implemented as distributed and parallel algorithm (i.g. Apache Spark MLlib implementation of Random Forest)
   
   - Scala and Apache Spark
        > Functional languages such as Scala allow us to represent mathematical concepts naturally, e.g. Probability Monad
        > Functional languages such as Scala, by promoting and encouraging immutability, are well suited to parallel or cluster computing,
          i.g., Apache Spark MLlib