Graphtools

Changes:

* added anisotropic kernel (alpha parameter from Coifman and Lafon 2006)
* added `to_igraph` function
* allowed `from_igraph` to return weighted kernel
* documentation updates

Changeset:
* significant changes to `MNNGraph` kernel creation: between batch connections are now downweighted relative to the sum of within-batch connections
* added fixed bandwidth kernel creation with `bandwidth` parameter
* exposed `clusters` attribute on `LandmarkGraph`
* added igraph->graphtools conversion function
* added graphtools->PyGSP conversion function
* `MNNGraph` parameter `gamma` renamed to `theta`

Changes:

* Uses tasklogger for logging
* Allows duplicates in building the graph; connections evaluated at nan are set to 1 with a warning.
* Landmarking is now done on the symmetric diffusion affinity matrix for true spectral clustering
* LandmarkGraph now exposes the transition matrix from samples to landmarks.

Graphtools 0.1.9.

Changes:
- `graphtools.base.Data` accepts `anndata.AnnData` input
- `graphtools.base.Data.inverse_transform` accepts column indices to return only a subset of ambient dimensions
- minor bugfixes and improvements to documentation

graphtools v0.1.7 implements minor bugfixes to handling of sparse matrices.

Other changes:
- `kernel_symm='none'` has been replaced with `kernel_symm=None`
- precomputed input matrix errors have been resolved
- additional testing of sparse matrix input has been added.

graphtools 0.1.4 greatly improves that handling of `pandas` and `scipy` sparse matrices.

Additional changes:
- all graphs accept gamma symmetrization
- unrecognised keyword arguments throw errors
- minor bugfixes