Energize Andover Summer 2017 Blog

Search results for

• Day 1

  Introduction

  . . . (read more)

  
  

• Day 2

  Initial distribution analysis

  . . . (read more)

  
  

• Day 3

  Night distribution continued

  . . . (read more)

  
  

• Day 4

  Median absolute deviations

  . . . (read more)

  
  

• Day 5

  Daytime usage first look

  . . . (read more)

  
  

• Day 6

  Time filtering

  . . . (read more)

  
  

• Day 7

  Generating complex time ranges with ical files, and improvements to time filtering

  . . . (read more)

  
  

• Day 8

  Debugging the calendar parser

  . . . (read more)

  
  

• Day 9

  Rolling windows intro

  . . . (read more)

  
  

• Day 10

  Rolling MAD, switching to time based windows

  . . . (read more)

  
  

• Day 11

  Centering time based windows, temperature data intro

  . . . (read more)

  
  

• Day 12

  Night data plotting, temperature data update

  . . . (read more)

  
  

• Day 13

  Percentage bounds and density plots

  . . . (read more)

  
  

• Day 14 - Summary

  Overview of my first couple of weeks, and a guide on how to make a blog

  . . . (read more)

  
  

• Day 15

  Team meeting, researching distribution functions and tests

  . . . (read more)

  
  

• Day 16

  Settling the null values and time gaps, looking at more filtered sample density estimations

  . . . (read more)

  
  

• Day 17

  Daily power trends and couple of stats test resources

  . . . (read more)

  
  

• Day 18

  Scipy, calculating the best fit distribution

  . . . (read more)

  
  

• Day 19

  Trapezoidal approximation of energy usage from power data

  . . . (read more)

  
  

• Day 20

  Team meeting with guest Viraj

  . . . (read more)

  
  

• Day 21

  Going over examples with the team

  . . . (read more)

  
  

• Day 22

  Working with the team (cont.)

  . . . (read more)

  
  

• Day 23

  Fitting the lognorm distribution

  . . . (read more)

  
  

• Day 24

  Abstracting the fit implementation, trouble with likelihood, and a way to compare quantiles

  . . . (read more)

  
  

• Day 25

  Plotting and comparing sample/model quartiles

  . . . (read more)

  
  

• Day 26

  Quantile translation for an adjusted data set

  . . . (read more)

  
  

• Day 27

  Tour of AHS, sample adjustment function and testing

  . . . (read more)

  
  

• Day 28

  More data to work with

  . . . (read more)

  
  

• Day 29

  Some Bancroft plots, time filter KeyError fix, and a promising lead on nonlinear regression

  . . . (read more)

  
  

• Day 30

  The holy grail of smart energy resources

  . . . (read more)

  
  

• Day 31

  Nonlinear regression lecture (polynomials, RBFs, kernels), testing out polynomial modeling on temperature data

  . . . (read more)

  
  

• Day 32

  Some remarks on least squares polynomial regression and an amazing machine learning library

  . . . (read more)

  
  

• Day 33

  Regression on higher dimensional input, 3D plotting with mplot3d

  . . . (read more)

  
  

• Day 34

  Regularized estimators

  . . . (read more)

  
  

• Day 35

  Distribution of the residuals, robust estimators, comprehending and implementing normalization

  . . . (read more)

  
  

• Day 36

  Fixing yesterday’s normalization misuse

  . . . (read more)

  
  

• Day 37

  Cross validation intro

  . . . (read more)

  
  

• Day 38

  Quick look at the cross validated residual plot

  . . . (read more)

  
  

• Day 39

  Research papers on power usage forecasting, residual plots of more data sets

  . . . (read more)

  
  

• Day 40

  Overview of forecasting methods, very promising results from multiple regression with lag features

  . . . (read more)

  
  

• Day 41

  Multi-step forecasting with multiple output regression, overview of the realtime anomaly detection research paper

  . . . (read more)

  
  

• Day 42

  Random forest vs ANN, forecast flow, more efficient window calculation with NumPy strides

  . . . (read more)

  
  

• Day 43

  Preparing the inputs/output matricies, testing the RF fit with median APE using different input/output/gap sizes, starting improvements to train speed

  . . . (read more)

  
  

• Day 44

  Index windows, more input features (holidays, descriptive stats, time data), trying non-default hyperparameters, flagging anomalies

  . . . (read more)

  
  

• Day 45

  Presentation, generalizing data frequency, quantile-based prediction intervals as threshold

  . . . (read more)

  
  

• Day 46

  Changing the prediction interval approach, starting abstraction of the training process

  . . . (read more)

  
  

• Day 47

  Starting on a model class

  . . . (read more)

  
  

• Day 48

  Adding extra features to the model, streamlining the training and prediction process

  . . . (read more)

  
  

• Day 49

  Getting prediction variances, auto-upsampling extra features, weighting samples by time, learning how to schedule commands

  . . . (read more)

  
  

• Day 50

  Restructuring the classes, predictions on multiple columns, outputting a CSV file, trial run

  . . . (read more)

  
  

• Day 51

  Slice based windows, more tests, bugfixing, writing up documentation

  . . . (read more)

  
  

• Day 52

  Bugfixing and class organization improvements

  . . . (read more)

  
  

• Day 53

  Multi-core processing for a performance boost

  . . . (read more)

  
  

• Day 54

  Logging the model properties, considering the implications of various energy savings quantification methods

  . . . (read more)

  
  

• Day 55

  Meeting, server setup intro

  . . . (read more)

  
  

• Day 56

  Slow progress with the setup

  . . . (read more)

  
  

• Day 57

  Getting the server working, starting to understand the mechanics

  . . . (read more)

  
  

• Day 58

  Future-proofing the energize module with a user guide and demo

  . . . (read more)

  
  

• Day 59

  Drafting a new data logging script, testing out APScheduler

  . . . (read more)

  
  

• Day 60

  Getting the team caught up, figuring out how to run the BACnet logging script

  . . . (read more)

  
  

• Day 61

  Successfully logging data, verifying compatibility with prediction process

  . . . (read more)