So apparently the repeat function did what I thought I wanted to do, so the
whole function reads:
basic_model = Model()
with basic_model:
# Priors for unknown model parameters
amplitude = Normal('amplitude', mu=0, sd=10000, shape = useToAs)
offset = Normal('offset', mu=0, sd=10000, shape = useToAs)
noise = HalfNormal('noise', sd=10000, shape = useToAs)
phase = Uniform('phase', lower = 0, upper = ReferencePeriod)
#parameters that define a two gaussian model
gsep = Savex[1]*ReferencePeriod/1024
g1width = Savex[2]*ReferencePeriod/1024
g2width = Savex[3]*ReferencePeriod/1024
g2amp = Savex[4]
Tg1width=theano.shared(g1width)
Tg2width =theano.shared(g2width)
Tg2amp=theano.shared(g2amp)
Tgsep=theano.shared(gsep)
#Calculate the X values for first gaussian, these have to wrap on a
period of [-ReferencePeriod/2, ReferencePeriod/2]
x = TFlatTimes - phase
x = ( x + ReferencePeriod/2) % (ReferencePeriod ) - ReferencePeriod/2
#Calculate the signal values for first gaussian
FlatS = np.exp(-0.5*(x)**2/Tg1width**2)
#Calculate the X values for second gaussian, these have to wrap on a
period of [-ReferencePeriod/2, ReferencePeriod/2]
x = TFlatTimes-phase-Tgsep
x = ( x + ReferencePeriod/2) % (ReferencePeriod ) - ReferencePeriod/2
#Calculate the signal values for second gaussian
FlatS += Tg2amp*np.exp(-0.5*(x)**2/Tg1width**2)
#Construct total vectors containing offsets, amplitude and noise
parameters.
#Each is (1024*useToAs) in length with the format (A_1, A_1, ...., A_1,
A_2, A_2, ..., A_2, ......, A_useToAs, ..A_useToAs)
NVec = theano.tensor.extra_ops.repeat(noise, 1024)
Offs = theano.tensor.extra_ops.repeat(offset, 1024)
Amps = theano.tensor.extra_ops.repeat(amplitude, 1024)
#combine total signal
Signal = Offs + Amps*FlatS
# Likelihood (sampling distribution) of observations
Y_obs = Normal('Y_obs', mu=Signal, sd=NVec, observed=FlatData)
Its still quite slow though.. i'm wondering if this is really the best way
to do it.
Cheers
Lindley
On Tuesday, July 12, 2016 at 5:53:45 PM UTC+1, Jesse Livezey wrote:
>
> Would tile then ravel work?
>
> http://deeplearning.net/software/theano/library/tensor/basic.html#theano.tensor.tile
>
> It looks like you're trying to turn a length n vector into a shape (n,
> 1024) matrix by repeating each element of the vector then raveling. If this
> is accuracte, tile should work for repeating.
>
> On Tuesday, July 12, 2016 at 12:17:07 AM UTC-7, Lindley Lentati wrote:
>>
>> Hi, I'm trying to implement an equivalent of the following:
>>
>>
>> NVec = np.array([np.ones(1024)*sig for sig in noise]).flatten()
>> Offs = np.array([np.ones(1024)*off for off in offset]).flatten()
>> Amps = np.array([np.ones(1024)*amp for amp in amplitude]).flatten()
>>
>> where noise, offset, and amplitude are theano vectors that contain free
>> parameters from a PyMC3 likelihood. When i try and do this I get the
>> following:
>>
>> ValueError: length not known: offset [id A]
>>
>> which I assume is because these vectors have no notion of length right?
>> I'm just wondering what the equivalent expression would be for use in
>> theano.
>>
>>
>> So far the closest i've gotten is creating a list of tensors:
>>
>> [theano.tensor.ones(1024)*noise[i] for i in range(useToAs)]
>>
>> but I cant seem to get them to concatenate into a single tensor.
>>
>>
>> If it helps, I know the length of everything before hand.. Apologies if
>> this is documented all over the place, I couldn't find an example that I
>> could translate into what I wanted to do.
>>
>>
>> Many Thanks
>>
>
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
