Monthly Downloads: 2
Programming language: Haskell
License: BSD 3-clause "New" or "Revised" License
Tags: AI     Machine Learning     Fei    

fei-nn alternatives and similar packages

Based on the "fei" category.
Alternatively, view fei-nn alternatives based on common mentions on social networks and blogs.

Do you think we are missing an alternative of fei-nn or a related project?

Add another 'fei' Package



This library builds a general neural network solver on top the MXNet raw c-apis and operators.


The Symbol API of MXNet synthesize a symbolic graph of the neural network. To solve such a graph, it is necessary to back every Symbol with two NDArray, one for forward propagation and one for backward propagation. By calling the API mxExecutorBind, the symbolic graph and backing NDArrays are bind together, producing an Executor. And with this executor, mxExecutorForward and mxExecutorBackward can run. By optimization, the the backing NDArrays of the neural network is updated in each iteration.


MXNet provides data iterators. And it can be wrapped in a Stream or Conduit. The mxnet-dataiter provides a implementation.

Some explanation

TrainM Monad

TrainM is simply a StateT monad, where the internal state is a store of all the backing NDArrays together with a Context (CPU/GPU). Both fit and forwardOnly must be run inside this monad.


initialize :: DType a => Symbol a -> Config a -> IO (Session a)

It takes the symbolic graph, and a configuration of 1) shapes of the placeholder in the training phase. 2) how to initialize the NDArrays

It returns a initial session, with it to run the training in TrainM Monad.


fit :: (DType a, MonadIO m, MonadThrow m, Optimizer opt) => opt a -> Symbol a -> M.HashMap String (NDArray a) -> TrainM a m ()

Given a optimizer, the symbolic graph, and feeding the placeholders, fit carries out a complete forward/backward phase, updating the NDArrays.


fitAndEval :: (DType a, MonadIO m, MonadThrow m, Optimizer opt, EvalMetricMethod mtr) => opt a -> Symbol a -> M.HashMap String (NDArray a) -> mtr a -> TrainM a m ()

Fit the neural network, and also record the evaluation.


forwardOnly :: (DType a, MonadIO m, MonadThrow m) => Symbol a -> M.HashMap String (Maybe (NDArray a)) -> TrainM a m [NDArray a]

Given a the symbolic graph, and feeding the placeholders (data with Just xx, and label with Nothing). forwardOnly carries out a forward phase only, returning the output of the neural network.


  • Please see the example in the examples directory.
  • Also see the examples of mxnet-examples repository.