Parameters setting

[JesseYANK]

Any suggestions for parameters setting. The performance of default setting is very poor

[jaredleekatzman]

Sorry for the confusion, the default parameters are place holders and are not meant to produce a working network.

I am not able to suggest parameters as optimal parameters are specific to whichever dataset you are using.

I suggest running a parameter search algorithm. For example use optunity for a random hyperparameter search.

Tue, Feb 21, 2017 at 3:46 AM JesseYANK [email protected] wrote:

Any suggestions for parameters setting. The performance of default setting is very poor

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jaredleekatzman/DeepSurv/issues/3, or mute the thread https://github.com/notifications/unsubscribe-auth/AFmmZs5En8D8mPopaNlrmx9opr9ibHQnks5reqRYgaJpZM4MHBmu .

[LeZhengThu]

@jaredleekatzman Hi, I tried your code and the network.predict_risk(X) gave output nan. Can you help explain why this happens? Thanks.

[jaredleekatzman]

@LeZhengThu could you provide more context? Did you train the network? Is it performing well on a validation set?

[JesseYANK]

@LeZhengThu

You should check your input data. If the some variables are extremely large when compared to other variables, you will get nan.

You can normalise the input data, which should work for your case.

On 10 Apr 2017, at 20:32, jaredleekatzman [email protected] wrote:

@LeZhengThu https://github.com/LeZhengThu could you provide more context? Did you train the network? Is it performing well on a validation set?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jaredleekatzman/DeepSurv/issues/3#issuecomment-292935641, or mute the thread https://github.com/notifications/unsubscribe-auth/ALq3gruouCCLFhgaxgRUtSVqM_eGojkyks5ruiFHgaJpZM4MHBmu.

[LeZhengThu]

@JesseYANK @jaredleekatzman Thank you for the reply. I'll try @JesseYANK method to normalise the data and get back to you guys.

[abracadabrac]

I had the same issue as LeZhengThu. I tried JessegThu method consisted is normalizing my data : x = ( x - mean(x) ) / 1000 It seems to works now.

[dareneiri]

@jaredleekatzman Can you please provide example of how you implemented Optunity for random hyperparameter search that is specific to deepsurv?

[jaredleekatzman]

Hi, I am currently working on releasing the Optunity script so that people can run their own hyper-parameter search. I hope to have it done soon.

On May 24, 2017, at 4:37 PM, dareneiri [email protected] wrote:

@jaredleekatzman https://github.com/jaredleekatzman Can you please provide example of how you implemented Optunity for random hyperparameter search that is specific to deepsurv?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jaredleekatzman/DeepSurv/issues/3#issuecomment-303844454, or mute the thread https://github.com/notifications/unsubscribe-auth/AFmmZgdozG2Pm0K-L1zaal_cjfh8OMieks5r9JUcgaJpZM4MHBmu.

[jaredleekatzman]

@dareneiri Sorry for the delay, but I've uploaded the script I used to run a random hyper-parameter search under the directory hyperparam_search/. If you still need to use it, let me know how it goes.

[hanxiaozhen2017]

@jaredleekatzman I will try the random hyper-parameter search and do we need to normalize our data before this step? Many thanks!