Python API

[colganwi]

@gillichu @uym2 thanks for creating this fantastic tool! It would be helpful to have a python API so lightweight inference tasks can be easily run in a jupyter notebook. I suggest taking a pd.DataFrame/np.array character_matrix and optionally a nx.DiGraph topology as input and outputting a nxDiGraph since this will enable compatibility with TreeData.

character_matrix = pd.DataFrame(character_matrix)
tdata = td.TreeData(obs = cell_metadata, obms = {"characters":character_matrix})
# With TreeData
tdata.obst["tree"] = laml.run_laml(tdata.obsm["characters"],depth_key = "time")
# Without TreeData
tree = laml.run_laml(character_matrix,depth_key = "time")

The inferred branch lengths and characters for ancestral nodes can be represented as node attributes in the nx.DiGraph. I typically use tree.nodes["0"]["time"] to store node times and tree.nodes["0"]["characters"] to store characters. When LAML is used for branch length estimation you could also take an nx.DiGraph as input then return a copy with "time" and "characters" attributes added. The key where these values are stored (e.g. "time") could be modifies with a parameter such as depth_key.

Happy to go into more detail here if that would be helpful. The goal is to have a simple API that works with or without TreeData.

[gillichu]

Hi William,

Thank you for creating this github issue, and for your interest in LAML! My apologies for the long silence on this issue. I do not have an update ready yet, but I plan to incorporate these changes in the next package update to LAML.

I will leave this issue open until I have resolved, but wanted to let you know this has not gone overlooked!