get different result using the same dataset and prototxt

[Jie Fang]

Recently, I wrote a python script to get precision and recall on validation set(val.txt). And of course, i got a result. After that  i run caffe to train the dataset again and didn't modify anything, then i also used that script to compute precision and recall, but i got a different result!

I trained it many times, and the results were different with each other. In other words, i got different parameters after each training. I think if i don't modify dataset and prototxt, i should get the same parameters, and therefore i should get the same precision and recall on validation set. Is it reasonable to get different result? Could anyone help me about this question?