Pseudo-absence generation

In this example we use a small synthetic dataset to show pseudo-absence generation in environmental space. Presences are restricted to warmer, lower-silicate conditions, and pseudo-absences are sampled outside the area of applicability.

Running the example

Before running the Python script we need to import the required packages and define the synthetic dataset. The candidate rows span temperature from 0 to 25 and silicate from 0.1 to 3. Observed presences are then selected from rows with temperature above 5 and silicate below 1.

Loading dependencies

import os
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from abil.pseudo_generation import generate_pseudo_absences

Generating synthetic data

species = "synthetic_species"
env_vars = ["temperature", "silicate"]
rng = np.random.default_rng(42)

missing_rows = pd.DataFrame(
    {
        "temperature": rng.uniform(0, 25, 1200),
        "silicate": rng.uniform(0.1, 3, 1200),
    }
)

presence_pool = missing_rows[
    (missing_rows["temperature"] > 5) & (missing_rows["silicate"] < 1)
]
merged_df = presence_pool.sample(n=150, random_state=42).copy()
merged_df[species] = 1
missing_rows = missing_rows.drop(index=merged_df.index)

Generating pseudo-absences

Next we call generate_pseudo_absences. With absence_ratio=1, the function targets one pseudo-absence for each observed presence. If there are fewer candidate rows outside the area of applicability, rows are sampled with replacement by default.

augmented = generate_pseudo_absences(
    merged_df,
    missing_rows,
    env_vars,
    [species],
    absence_ratio=1,
    min_presence=50,
)

Plotting

Now that we have pseudo-absences we can plot them in environmental space:

pseudo_absences = augmented[augmented[species] == 0]

fig, ax = plt.subplots(figsize=(7, 5))
ax.scatter(
    missing_rows["temperature"],
    missing_rows["silicate"],
    alpha=0.15,
    s=12,
    label="Candidate rows",
)
ax.scatter(
    merged_df["temperature"],
    merged_df["silicate"],
    marker="x",
    s=28,
    label="Observed presences",
)
ax.scatter(
    pseudo_absences["temperature"],
    pseudo_absences["silicate"],
    alpha=0.85,
    s=45,
    label="Pseudo-absences",
)
ax.set_xlabel("temperature")
ax.set_ylabel("silicate")
ax.set_title("Pseudo-absences from synthetic data")
ax.legend()
fig.tight_layout()
plt.savefig("pseudo_absence.png", dpi=300, bbox_inches="tight", facecolor="white")
plt.show()