I’ve trained a machine learning model in Orange that classifies dogs and cats with great accuracy. However, when I export the model to a pickle file and load it in Python, it consistently predicts “cat” regardless of the input data.

This is what i wrote in python:

import pickle
from PIL import Image
import numpy as np

modello = 'modelli/catDogsLogisticRegression.pkcls'

def load_model_from_pickle(modello):
    try:
        with open(modello, 'rb') as file_pickle:
            model = pickle.load(file_pickle)
            return model
    except FileNotFoundError:
        print(f"File {modello} non trovato.")
        return None

def preprocess_image(image_path):
    # Carica l'immagine
    img = Image.open(image_path)
    # Ridimensiona l'immagine e convertila in scala di grigi
    img = img.resize((32, 64)).convert('L')
    # Converte l'immagine in un array numpy e ridimensiona in un unico vettore
    img_array = np.array(img).reshape(1, -1)
    return img_array

# Esempio di utilizzo
loaded_model = load_model_from_pickle(modello)
if loaded_model:
    print("Modello caricato con successo")
    # Utilizzo del modello
    # Carica e pre-elabora un'immagine
    image_path = 'cane.jpg'
    new_data = preprocess_image(image_path)
    # Prevedere la classe del nuovo esempio
    predicted_class = loaded_model.predict(new_data)[0]
    print("Classe prevista:", 'Gatto' if predicted_class == 0 else 'Cane')
else:
    print("Errore nel caricamento del modello.")

Any ideas why the model behaves differently in Python? How can I fix this issue?