From 0d69176691bee511061e29d970481989b10d9488 Mon Sep 17 00:00:00 2001 From: Hillary Scannell Date: Fri, 27 Jul 2018 16:21:57 -0700 Subject: [PATCH] Update README.md --- README.md | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index cf8da5c..439efb4 100644 --- a/README.md +++ b/README.md @@ -16,9 +16,20 @@ This project involves tracking extreme and prolonged warming events in sea surfa ![Location of Initial time series](https://github.com/hscannell/MHWpredict/blob/master/data/datamap.png) *** -| Leader Board | | -| Model | Description | Score | -|--------------|-------------|-------| +### Leader Board +Average score across all four time series (30n120w, 30n140w, 40n160w, & 50n140w) + +| Model | Description | Error | +|---|---|---| +| TBATS | Exponential smothing state space model with Box-Cox Transformation | ? | +| AR(6) | AutoRegressive order 6 models | MSE = 0.10 | +| ARMA | AutoRegressive Moving Average, order(6,2) | MSE = 0.13 | +| SARIMAX | Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors, order(6,2,1)(6,2,1,4) | MSE = 0.13 | +| Persistence | "walk-forward" validation | MSE = 0.28 (1-day), 0.29 (3-day), 0.65 (9-day) | + +| Long Short-Term Memory (LSTM) | Architecture | Error | +|---|---|---| +|LSTM_1var | Sequential 5 neuron LSTM, 1 neuron Dense output layer, tanh activation, SGD optimizer, fit with 20 epochs, 520 batch size, forecasting 5-day sequence based on the previous 20 days, features are only SST | |