Comparison to TensorFlow

[kp...@cs.washington.edu]

I couldn't find a direct comparison of TensorFlow and Ceres-Solver online so I'm posting to start a discussion. TensorFlow is supposed to be a general machine learning toolkit so I am wondering what the exact differences might be.

Here are some I've determined:

• TensorFlow does not have implementations for certain optimization methods such as Levenberg–Marquardt or BFGS.
• Ceres doesn't do backwards auto-differentiation which means it isn't suited for neural network style problems.
• Ceres supports local parameterization e.g. for parameters on non-Euclidean manifolds.

TensorFlow comes with a convenient Python interface so I'm wondering whether there is a clear choice between the two when one's dealing with simple parametric optimizations (e.g. estimating parameters of a 10 parameter non-linear equation from data.)

[Sameer Agarwal]

For small problems of the kind you are talking about, the autodiff performance does not matter.  Tensor flow is gradient descent and it's various variants which work well with a lot of data and when the exact local optimum does not matter.

For your problems use Ceres.

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/b6cdef0c-4cd5-47ed-bb1e-720866518865%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.