as far as i try all data that model take are fixed size or fixed shape, i had problem that i give data to model but it has no fixed shape and i can not cut part of it as all of it are important,so can i make model that have unshaped data as input then take unshaped data to predict .. like have small problem that can be represented by N like famous problem travel sales man as in it N refer to number of cities
The first thing, which comes to mind speaking about unfixed input, is a recurrent neural network. Such a model adresses dynamic behavior. Maybe, it's possible to represent data as a sequence? I don't know nature and specificity of your data, but if each instance in the dataset consists of a different number of some "similar events", it might be reasonable to think in this direction.
The other way which might help is to compute statistics on the data. Then it will be another feature space, but in some cases it could be enough to achieve high accuracy of predictions. For example if there are k features of the same nature (say, temperature in different days), then average, standard deviation, min, max, etc. could be considered. And statistics don't depend on the length of each instance in the dataset, so this number k could differ.