aplowman / spatial-sites Goto Github PK

Sites.sites is too confusing.

s1 = Sites(coords=..., labels=...)

After instantiation, s1.coords[0] == s1.x

Currently, we can reassign the default coordinate attributes like: s1.x = 0, after which s1.x no longer refers to the same thing as s1.coords[0]. So to avoid this, prevent reassignment of s1.x and similar attributes.

For Sites objects: s1, s2:

Addition-like

Add a vector to all site coordinates (vector must have correct shape):

• s1 += vector
• s1' = s1 + vector == vector + s1
• s1.translate(vector)

Combining multiple Sites objects to increase the number of coordinates (all Sites objects must have the same labels defined):

• s1 += s2
• s1' = s1 + s2 != s2 + s1
• s1' = Sites.concatenate([s1, s2, ...])

Subtraction-like

Subtract a vector from all site coordinates (vector must have correct shape):

• s1 -= vector
• s1' = s1 - vector != vector - s1
• s1.translate(-vector)

Removing coordinates according to some condition(s):

• Remove coordinates based on a label condition: s1.remove(species='Zr')
• Remove coordinates based on coordinate positions s1.remove(s1.x > 0.5)

Multiplication-like

Scaling coordinates by a scalar number:

• s1' = s1 * number == number * s1
• s1 *= number

Transform coordinates according to a linear operator (matrix), mat (what are the conditions on mat with respect to the dimension and basis attributes?):

• Allowed only if vector_direction == column:
  • s1' = mat @ s1
• Allowed only if vector_direction == row:
  • s1' = s1 @ mat
  • s1 @= mat
• Some form of outer product multiplication; e.g. applying one set of sites at each site in another set (would need to think about bases). Would this use __mul__ or __matmul__? This became s1.tile(s2).

Division-like

Scaling coordinates by a scalar number (number must be non-zero):

• s1' = s1 / number
• s1 /= number

Get a subset of coordinates

• Get coordinates based on label value: coords = s1.whose(label_name=label_value)
  • Note s1.whose(label_name=label_value) is the same as s1.sites[:, s.label_name== label_value] (if vector_direction == column)
• Get coordinate indices based on label value coords_idx = s1.index(label_name=label_value)
• Get coordinates based on coordinate positions coords = s1.where(s1.x <= 0.1)

Comparison

• Equality (==) should check coordinates and labels are equivalent (or to within a small tolerance)
• Other comparison operators (<, <=, >, >=):
  • Between two Sites, just pass directly to the corresponding Numpy methods
  • Between a Sites and a number or array, compare the self.coords with the number/array

Properties

• Geometric centre: s1.centroid
• Bounding box (orthogonal; assuming standard basis): s1.bounding_box

Add/remove label objects

• s1.add_labels(label_name_1=(..., ...), label_name_2=(..., ...), ...)
• s1.remove_labels('label_name_1', 'label_name_2', ...)

Other

• Copy: s1' = s1.copy()
• Rotate about point: s1.rotate(origin, [rot_mat, quaternion])
• Get all x/y/z/nth coordinate components: s1.get_components(component_idx)
• Add x/y/z attributes

Each index into a Sites object should return a SingleSite object with similar attributes to the parent object.

Consider a sites object:

s1 = Sites(coords=..., labels=...)`

We should be able to make a new sites object that contains a subset of the sites in the original object by using an integer or boolean index array:

s2 = s1[[1, 7, 2]]

I don't think there is a need for multi-dimensional indexing in this manner.