Baiz

challenge1/analysis/baiz/README.md

@@ -0,0 +1,57 @@
+# Overview of Challenge Results
+
+This readme file provides a high level description of all the work performed during this 
+challenge. The data science process starts by a traditional EDA (Exploratory Data
+Analysis) using Pandas Profile and other traditional visualisation techniques. 
+Then the work moves towards actual prediction of EnergyProduction for June 2013. 
+
+## Short Files Description Overview
+
+All files added were only added in the present directory, and these are:
+
+- README.md (This file)
+
+- requirements.txt (Python requirements file to reproduce venv)
+
+- 001_ExploratoryDataAnalysis.ipynb (Notebook to Explore DataSet)
+
+- 001_ExploratoryDataAnalysis.pdf (PDF of Notebook to Explore DataSet)
+
+- data_profile_<dataset>.html (HTML with data profiling report)
+
+- 002_VAR.ipynb (Notebook that performs predictions)
+
+- 002_VAR.pdf (PDF of Notebook that performs predictions)
+
+- predicted_energy_production.csv (Requested File)
+
+- mape.txt (Requested File)
+
+
+# Exploratory Data Analysis
+
+The exploration process revealed a very good quality dataset with key features 
+for a Machine Learning model, these were clearly: 'House', 'Temperature', 'Daylight', 'EnergyProduction' (and clearly the timestamp). 
+
+Look at file 001_ExploratoryDataAnalysis for more details. 
+
+
+# Predictions/Forecasting
+
+When considering forecasting of multivariate timeseries, few techniques could come to mind, but officially 
+one techniques is particularly used: Vector Autoregressions (https://www.statsmodels.org/dev/vector_ar.html).
+
+More details about the exact implementation can be see in the file 002_VAR. 
+
+
+# Future Work
+
+An attempt was also made to perform some RNN LSTM as this type of NN are now demonstrating great performance 
+for time series analysis. Unfortunately, not enough time was available to obtain results. In this regard one
+aspects that was particularly considered was the use of not just RNN-LSTM but also their interpretation using 
+SHAP. This is still an area of active research and hopefully in the near future this will be easily available. 
+
+More on SHAP for RNN is available here: 
+
+- https://github.com/slundberg/shap/issues/213
+- https://towardsdatascience.com/interpreting-recurrent-neural-networks-on-multivariate-time-series-ebec0edb8f5a

challenge1/analysis/baiz/mape.txt

@@ -0,0 +1 @@
+4.023386014469117