[GSoC] Set up HDF5 file for element's periodic data

[gabrielasd]

Translate the current periodic module code and associated CSV files into a HDF5 structure that can be accessed via Species class @Property methods.

Modules associated with this change:
periodic.py
species.py

Data files associated with change:
data csv
data provenance & descrption

Example of current Species property method affected by change:

AtomDB/atomdb/species.py

Lines 472 to 482 in 464f116

	def atmass(self):
	r"""Atomic mass in atomic units.
	Returns
	-------
	atmass : dict
	Two options are available: the isotopically averaged mass 'stb', and the mass of the most
	common isotope 'nist'. For references corresponding to each type check
	https://github.com/theochem/AtomDB/blob/master/atomdb/data/data_info.csv
	"""

Tasks

• Port information in csv files to HDF5 format (e.g., atomic number, symbol, covalent radius, etc.). This step may involve deciding a data schema for the HDF5 file and writing a script to port the data.
• Refactor the periodic module
• Refactor Species to load periodic data from HDF5;relevant function:
  

  AtomDB/atomdb/species.py

  Line 65 in 464f116

  def scalar(method):

  )
• Fix periodic test module

[enjyashraf18]

Just wanted to share some progress and ask a couple of things.

So for data, props, srcs, units, prop2col, num2str, and str2num from data_info:
Before, data was just about reading the CSV file and now that I’m switching to HDF5, I’m not exactly sure what to do with it.
I managed to read the rest from the HDF5 file though and I’m still working on simplifying that part and making sure the output matches the old one exactly.

Also (just to double-check) I know that our plan for now is to keep the same structure, so most of setup_element will stay unchanged for now right?

[gabrielasd]

Hi @enjyashraf18 thanks for the update.
Yes, the idea is to preserve the structure (in particular the existence of the Element, element_{number, symbol, name} methods) but if the particulars of how the class is instanced need to change I have no problems with that.

I can give you an immediate solution right now, but I'll go over your changes and an idea for specific changes to the periodic module before our meeting tomorrow. At the latest time, we can discuss this tomorrow.

@msricher if you have capacity before me and can help Enjy that's fine too.

[msricher]

Hi, can you be more specific about what you're doing? I'm happy to help for now, just need more context.

[msricher]

@enjyashraf18 I think I know what you are asking about, now. For use of the AtomDB library, you just need to load the Elements HDF5 file into memory as a global variable (probably in a class definition as a static property). Then, when we have a Species instance and want its covalent radius, for example, we just return the correct value from the global HDF5 data.

EDIT/clarification: Most species properties will be specific to that species (Z, charge, spin, n_exc.), but the Elements data is common to all species of an atomic number Z so we just use an indirection to obtain it from a global table.

[enjyashraf18]

Thank you @gabrielasd and @msricher
My question is about data, when it’s returned from the get_data, it was previously just the result of reading the CSV file (check this line), and I think its usage is mentioned in the docstring.
Now after refactoring get_data to use the HDF5 file, I was able to refactor props, srcs, units, prop2col, num2str, and str2num, but I’m still not sure what to do with data

[msricher]

For this task, just save the HDF file of all of the data. So for each property, store the vector of 118 values, one for each element.

[gabrielasd]

Hi Enjy, it is OK to get rid of that data variable. Like Michelle is pointing out we can get the property values directly from the loaded pytables object.

With respect to the setup_element function we, can keep it as is, as long as trying to adapt it to the new form of accesing data is simpler that just implementing things differently.

Going with the initial attempt to keep the current code structure, and based on what you alredy have, what would need to change is this function:

AtomDB/atomdb/periodic.py

Line 160 in 464f116

f = make_property(data, prop, prop2col)

and the definition of variable type for the doctrings of the atomic properties:

AtomDB/atomdb/periodic.py

Line 135 in 464f116

t = type(data[0][prop2col[prop][""]]).__name__