From cff31dd419c9e2da999afbe16320fd7f62c5f0ef Mon Sep 17 00:00:00 2001 From: Aleksander Date: Thu, 23 Nov 2017 14:06:46 +0100 Subject: [PATCH] Update README.md --- README.md | 19 +------------------ 1 file changed, 1 insertion(+), 18 deletions(-) diff --git a/README.md b/README.md index 1e98c57..11b9578 100644 --- a/README.md +++ b/README.md @@ -41,7 +41,7 @@ y_pred,std=gp.predict(x_test) ``` The example creates a mapping z(x) where both x and z are 1d vectors using a neural network with 1 hidden layer. -The CovMat layer creates a covariance matrix from z using the covariance function x\*exp(-0.5*|z1-z2|**2) with noise y where x and y are learned during training. +The CovMat layer creates a covariance matrix from z using the covariance function v\*exp(-0.5*|z1-z2|**2) with noise y where x and y are learned during training. x and y are available after training as gp.layers[-1].var and gp.layers[-1].s_alpha. The gp.fast_forward() function can be used to extract the z(x) function (It skips the last layer that makes an array of size [batch_size, batch_size]). @@ -71,20 +71,3 @@ DKL Prediction:

We see that DKL solves the problem quite nicely, given the limited data. We also see that for x<-0.5 the std.dev of the DKL model does not capture the prediction error. - -### Prediction error on MNIST - -The example_mnist.py script tries takes a crack at the classic MNIST classification problem (even though we currently are using a regression algorihm), to see if we are able to predict when we are making an error. -Since no large-scale GP algorithms are implemented yet, we have to limit training data. We use mini-batch learning on the log marginal likelihood to find z(x). -Then fit a GP model on 5000 MNIST training samples. - -Also, note that we are using regression on the labels. There is no support for real classification yet. However, despite these concerns, the results are decent. - -The following plots all the testing samples according to predicted std.error: - -

-

- - -
We seem to miss when the predicted std.dev is high.
-

\ No newline at end of file