Python classification (descriptor)

Available from Breeze v2026.2

On a Windows machine, you need to install Python version 3 or higher and the py Python launcher, or specify a custom python interpreter

First select Descriptor as node type and then select Python script as the method. Select the python script file using the Browse button or type the path to the script file directly.

The type value must be 1-based as 0 is reserved in Breeze for NO_CLASS

Parameters

Python file

Select the python file.

Python interpreter optional

Use the file selector to specify a custom Python executable. If left empty, the system will attempt to use:

  1. The global system command (py on Windows, python3 on Linux).

  2. The bundled Breeze Python environment (used as a fallback if the global command is not found).

Apply to

  • Object

    • Applies the python file content on the objects.

  • Pixels

    • Applies the python file content on the pixels.

Names file

Newline separated classes/labels file

Almonds
Shell
Hazelnut
Pecan
Walnut

Example for nuts.names

If no names file is supplied the type will only be a numerical value.

Pretreatments

See Pretreatments

Output

  • Value

    • A float32 value from the script

    • Could be a confidence value from the classification

  • Type

    • A int32 value, corresponding to the same row in the Names file.

Example

Below is a high-performance classification script utilizing a custom Neural Network forward pass via NumPy. It expects an uncompressed .npz file containing the weights and biases alongside the script. Based on the Classification of nuts step 3 data.

Training

Python
import pandas as pd
import numpy as np
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import classification_report

df_train = pd.read_csv("nuts_train.csv", sep=";")
X_train = df_train.iloc[:, 1:].values
y_train = df_train.iloc[:, 0].values

df_test = pd.read_csv("nuts_test.csv", sep=";")
X_test = df_test.iloc[:, 1:].values
y_test = df_test.iloc[:, 0].values

model = MLPClassifier(hidden_layer_sizes=(64,), activation='relu', max_iter=1000, random_state=42)
model.fit(X_train, y_train)

y_pred = model.predict(X_test)
print(classification_report(y_test, y_pred))

np.savez("nn_weights.npz",
         W1=model.coefs_[0].astype(np.float32),
         b1=model.intercepts_[0].astype(np.float32),
         W2=model.coefs_[1].astype(np.float32),
         b2=model.intercepts_[1].astype(np.float32))

Classifier

Python
import os
import numpy as np

W1, b1, W2, b2 = None, None, None, None


def setup():
    global W1, b1, W2, b2

    script_dir = os.path.dirname(os.path.abspath(__file__))
    weights_path = os.path.join(script_dir, "nn_weights.npz")

    data = np.load(weights_path)
    W1 = data['W1'].astype(np.float32)
    b1 = data['b1'].astype(np.float32)
    W2 = data['W2'].astype(np.float32)
    b2 = data['b2'].astype(np.float32)


def perform(frame_data, num_bands):
    pixel_count = len(frame_data) // num_bands
    if pixel_count == 0:
        return []

    x = np.array(frame_data, dtype=np.float32).reshape((pixel_count, num_bands))

    z1 = np.dot(x, W1) + b1
    a1 = np.maximum(0, z1)
    z2 = np.dot(a1, W2) + b2

    results = np.empty((pixel_count, 2), dtype=np.float32)
    results[:, 0] = np.max(z2, axis=1)
    results[:, 1] = np.argmax(z2, axis=1) + 1

    return results