okreshma
(Reshma O.K.)
August 21, 2019, 5:25am
1
I am trying to implement two similar network models with two LSTM layers whose parameters are not tied.
For eg., If I create two instances of the same model as:
model1 = Model()
model2 = Model()
Would the parameters of the above networks be tied? Kindly suggest a way to solve this.
JTiC
August 21, 2019, 8:11am
2
They wouldn’t!!!(for 20 characters.)
okreshma
(Reshma O.K.)
August 21, 2019, 9:35am
3
Thank you @JTiC .
So the parameters won’t be tied, but I didn’t understand why it is specified as 20 characters.
okreshma
(Reshma O.K.)
August 21, 2019, 1:19pm
5
I tried comparing the parameters of the two models with the function described in the following post.
check if models have same weights
And I got the result as “mismatch”. The comparison of the parameters was done after inference and before training. Is this the proper method to check if the parameters are tied or not? Kindly suggest the correct approach for the same.