I think common things in documentation should be written in a consistent way. Right now I can think of the following points:

1. Whether to name entities which can be transformed to ndarray as array or array-like.

2. How to list different input options. Should curly braces always be used? For example:

   X : {array-like, sparse matrix}
   

3. Optional arguments / default values. Some convention should be adopted, like:

   kernel : string, default ’rbf’
   

   or

   kernel : string (default='rbf')
   

   The word optional should be dropped anyway as redundant.

4. How to name X and y arrays in fit and predict methods. The one adopted convention would be helpful.

5. How to mention shapes of arrays. I was told that X, shape (n_samples, n_features) was adopted, is it true? How about 1-d arrays: (n_samples) or (n_samples,)?

Anyway, do you think it is an issue at all? Maybe core developers could decide and update coding guidelines? I listed only some points.