Fixing random seed and getting same score for all the different values of parameters

I want to train an autoencoder but I found that the score changes every time I run the code for the same parameters . (so it is difficult to document the results in a paper/thesis .. because for same parameters values , the score is 0.11 for a run and 0.55 for another run! )

So I fixed the random seed to 99 or example .

Now ,the problem is that when I change the parameters (dropout , learning rate ,number of epochs and batches ), the score is fixed for different values of parameters!!!

The score changes only when I change the number of hidden units or layers. I do not understand why !

I want to get the same score if I use the same values of parameters , not for any values of parameters !

I am using Tensorflow .

For the weights of the autoencoder , I used tf.Variable(tf.random_uniform([shape],seed=99)).