[SciPy-User] [Numpy-discussion] Weibull analysis ?

[josef.pktd at gmail.com]

On Tue, Nov 30, 2010 at 10:55 AM, Skipper Seabold <jsseabold at gmail.com> wrote:
> On Mon, Nov 29, 2010 at 12:53 PM, David Trémouilles
> <david.trem at gmail.com> wrote:
>> Thanks for this starting point Skipper !
>> What you mentioned is a small part of what I'm looking for.
>>
>
> I don't know of anything that can do these things in Python (that
> doesn't mean anything though).  A brief look through the following
> references, I don't see anything that couldn't be accomplished with
> scipy.  You can look to the statsmodels scikit if you want some
> structure.  Please post your code, if you get any further on this.
>
> References in-lined for my own edification.
>
>> Among other feature regarding Weibull analysis I'm interested in:
>>   - Type 1 right censored data Maximum likelihood estimator
>
> http://www.weibull.com/LifeDataWeb/analysis_parameter_methods.htm#suspended_data
>
>>   - Fisher matrix for confidence bound
>
> http://www.weibull.com/LifeDataWeb/fisher_matrix_confidence_bounds.htm
>
>>   - Likelihood ratio bound
>
> http://www.weibull.com/LifeDataWeb/likelihood_ratio_confidence_bounds.htm
>
>>   - Parameter estimation of mixed weibull models
>>   - ...
>
> http://www.weibull.com/LifeDataWeb/the_mixed_weibull_distribution.htm#parameters

Thanks Skipper, nice references.

Per Brodtkorb still has the best code for this that I have seen
http://code.google.com/p/pywafo/

I haven't managed to work my way through profile likelihood yet. With
generic mle it should be 20 lines of code or less to get mle estimate
and parameter_covariance estimates. Estimating the lower bound in a 3
parameter weibull might have problems with mle. Per has Maximum
Product Spacings as alternative estimator. (I'm using generalized
method of moments with quantile matching as an alternative.)

I haven't seen anything on mixture modeling in Python other than
gaussian. If there are only a few mixtures, then mle should be able to
handle it without having to use an EM algorithm.

(generic survival/hazard/failure model estimation with censored or
binned data is on my plans for statsmodels, I just need to put the
pieces together.)

Josef

>
> Skipper
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>