mahynski/slovenian-site-strawberry

Summary

Refer to this manuscript for details about training and model backgrounds. You can also refer to the associated GitHub repo. The figures given here summarizes the models trained and their expected performance (error bars below are 1 standard deviation). Models in this repo are labeled by model{name}[metric].pkl, where metric refers to the performance metric used to evaluate and optimize the model (either total efficiency, teff, or accuracy, acc). In the figures at the right, the performance metric is reported in parentheses. These performances and uncertainties are intended to reflect the expected performance on new data.

There are two variations of models reported here: one which uses Phosphorus as a feature and one which excludes it. Models which do not accept Phosphorus have a -no-p suffix to their name, e.g., "model-dt-stump-no-p[acc].pkl". The figure on top reflects the performance of models WITH P, while the figure on the bottom reflects the performance of models WITHOUT P (-no-p suffix).

Warnings

These models were generated as part of an academic exercise and are not intended for a production environment. These are provided "as-is" with no warranty of any kind. Refer to the license for more information.

Inputs

The inputs are stable isotope and trace element measurements in the following order. For models which do not use Phosphorus, simply remove it to maintain the correct ordering.

SITE	Units
18O	per mille
13C	per mille
15N	per mille
34S	per mille
Na	micrograms/gram
Mg	milligrams/gram
Al	micrograms/gram
P	milligrams/gram
S	milligrams/gram
K	milligrams/gram
Ca	milligrams/gram
V	nanograms/gram
Mn	micrograms/gram
Fe	micrograms/gram
Co	nanograms/gram
Ni	nanograms/gram
Cu	micrograms/gram
Zn	micrograms/gram
As	nanograms/gram
Rb	micrograms/gram
Sr	micrograms/gram
Mo	nanograms/gram
Cd	nanograms/gram
Cs	nanograms/gram
Ba	micrograms/gram

These units should follow the convention of the models' training set, datasets/mahynski/slovenian-site-strawberry.

Usage

import joblib
from huggingface_hub import hf_hub_download

model = joblib.load(
  hf_hub_download(
    repo_id="mahynski/slovenian-site-strawberry",
    filename="model-dt-stump-no-p[acc].pkl", # Select one of the models in this repo
  )
)

model.predict(X)

Or you can use the PyChemAuth library.

import pychemauth

model = pychemauth.utils.HuggingFace.from_pretrained(
  model_id="mahynski/slovenian-site-strawberry",
  filename="model-dt-stump-no-p[acc].pkl",
)

model.predict(X)