I have a few points of feedback after working across different dask collections for benchmarks and examples.

Currently, dask collections have a partitioning argument (depending on the particular method being used) and an attribute to show the number of partitions:

1. When moving between different from_* methods in arrays and bags/dataframes, I often pause to remember which argument is used to specify the number of partitions or chunk sizes. Thoughts on making the number of partitions a consistent argument (ideally, partitions=X or npartitions=X) between all collections? For arrays, I almost always find myself inverting for the chunk sizes to arrive at a particular number of partitions (especially when I'm using the distributed scheduler). It would be comforting to know that I can create a collection and always use a partitions or npartitions argument whether I'm using dask.bag.from_filenames or dask.array, any reason this might not be feasible for all collections?

2. Assuming Blaze slicing #1 is a valid point, thoughts on making the argument to specify partitions the same as the attribute for querying the number of partitions? For dask.array, I need to remember chunks vs. numblocks.

3. It might be handy to show the number of partitions in the repr for each collection, e.g.:

>>> x
dask.array<arange-..., shape=(100,), dtype=int64, chunksize=(2,), partitions=10>

>>> y
dd.Series<from-dask-arraycdefdd873cdf06131bcc8da67a0e65fb, divisions=(0, 2, 4, ..., 98, 99), partitions=10>

>>> b
<dask.bag.core.Bag at 0x11312e0f0, partitions=10>