No, you cannot do this.
Your series of (mostly duplicated) questions along these lines has assumed that you can generalize from a finite set of sample values to get "the" formula for the function and use 'the" formula in calculate the function composition. This is a false premise. There are (literally) an infinite number of different functions that fit any finite set of sample values and there is no way of distinguishing between the "correctness" of the members of that infinite set. The "real" function might oscillate 319 times between each of the sample values that you know, but unless you know about the precise form of that oscillation you cannot do function composition except right at the exact points you know.
Assuming linearity or cubic fitting between the points you know is mathematically unsupportable in the general case -- it doesn't even work for the sin() function you mentioned as an example in your other copies of this thread.