Replay-identification

+ Fix track labels
+ Set likelihood to zero not on track

+ Add w-track specific 1D random walk state transition
+ Add random walk state transition
+ Make interface more consistent with scikit-learn

+ Minor fix to handle the case where the GLM does not converge in the spiking model

+ Implement smoother, which incorporates both past and future information (feb1db0)
+ Allow user to specify `lfp_model`, which can be any `scikit-learn` kernel density estimator (5309783)
+ Raise NotImplemented Errors for loading and saving models (c8bcc93)
+ Convert all data to numpy arrays to handle Pandas DataFrames (a4f58c0)
+ Added some model checking plots (ecc8d4c)
+ Remove speed from default likelihood (37a6d6a)

+ Fix handling of NaN
+ Don't estimate the movement variance
+ Fix the predict function
+ Fix the default penalty and knot spacing for spiking model
+ Add option to specify bin number

+ Make speed knot selection for replay state transition based on the data or user specified.