Multiple inputs & nearest neighbors #7
Conversation
| density=4.7 | ||
| prop="band_gap" | ||
|
|
||
| output = exact_match.query_material_property(formula, prop, one_only=False) |
There was a problem hiding this comment.
This has the potential to fail in future as new materials are added to materials project.
Maybe assert len(output) >= 9 is a better test
| docs = [doc for doc in docs if doc.symmetry.crystal_system.value == lattice] | ||
| return docs | ||
|
|
||
| def query_material_property(self, formula, desired_prop, space_group_num=None, density=None, lattice=None, one_only=True): |
There was a problem hiding this comment.
I am still a little confused about how it'll be implemented in the agent. Is the agent supposed to decide if they will query one_only or not?
If the user/agent should have access to it, the warning of returning one_only should be in this function. If this will be a private function and the control will be done in the get_prop function, then we don't need the one_only argument here
| all_values = self._get_prop(all_values, property) | ||
| return sum(all_values) / len(all_values) | ||
|
|
||
| def k_vals(self, input_param:dict, k:int): |
There was a problem hiding this comment.
I'd change this fxn name for something like has_k_vals. It took me a while to understand it was checking if there's enough examples to fill the context already
| all_props = self._get_prop(docs, desired_prop) | ||
| if not docs: | ||
| print ("Warning: no matches found. Returning the average of closet matches.") | ||
| nearest_neighbor_avg = self.get_weighted_average_of_neighbors({"formula": formula, "desired_prop": desired_prop, "space_group_num": space_group_num, "density": density, "lattice": lattice}) |
There was a problem hiding this comment.
We need to compare this average pooling scheme with the ICL to decide which one to keep. But I think it's ok to keep that for now
In this PR:
Tools can now accept the following parameters:
any of the non-required parameters are left out of the query
Alternative to ICL:
I added a demo notebook for this as well