Question about the problem encountered in "Cubic Crystal Test.ipynb" under the "examples" folder in the original M3GNet

[Claudia-Hello]

Could I ask a question that is not in the Matbench framework? It's a question related to the original M3GNet model on https://github.com/materialsvirtuallab/m3gnet GitHub.

In M3GNet, in the file examples/Cubic Crystal Test.ipynb

data = pd.read_html("http://en.wikipedia.org/wiki/Lattice_constant")[0]
data = data["Crystal structure"][0].
    ~data["Crystal structure"].isin(
        ["Hexagonal", "Wurtzite", "Wurtzite (HCP)", "Orthorombic", "Tetragonal perovskite", "Orthorhombic perovskite"]
    )
]
data.rename(columns={"Lattice constant (Å)": "a (Å)"}, inplace=True)
data.drop(columns=["Ref."], inplace=True)
data["a (Å)"] = data["a (Å)"].map(float)
data = data[["Material", "Crystal structure", "a (Å)"]]
data = data[data["Material"] ! = "NC0.99"]

In the code above:

data["a (Å)"] = data["a (Å)"].map(float)

This line of statement is to convert the string type to float type. However, the column Lattice constant (Å) (that is, the column a (Å)) in the first table in this url: https://en.wikipedia.org/wiki/Lattice_constant includes multiple string formats, such as 3.567, a = 3.533 c = 5.693, and a = 5.27 b = 5.275 c = 7.464. What should be done about the case where more than one value is included? If we map it to float type directly, then an error will occur.

Thank you!

[janosh]

in the future, such questions are better asked on stackoverflow.

you can pass any custom function to pd.DataFrame.map(). here's an example of converting the first value in each string. you could change the function to return tuples with all floats instead.

import re
import pandas as pd

data = pd.read_html("http://en.wikipedia.org/wiki/Lattice_constant")[0]

def extract_first_float(value: str) -> float:
    try:
        return float(value)
    except ValueError:
        match = re.search(r"[-+]?\d*\.\d+|\d+", value)
        if match:
            return float(match.group())
        else:
            return float("nan")

data["a (Å)"] = data["a (Å)"].map(extract_first_float)

[Claudia-Hello]

Thanks for the reply, I will ask this kind of question on Stack Overflow in the future.
Yes I just customized a function to extract the first value, but my question is: should only the first float number be read in this case, or are both (or all 3) values to be read?
Thank you!

[janosh]

i don't know what your goal is but in most cases you'd want all 3 values

[Claudia-Hello]

Ah ok I understand that, thank you so much!