Use a pretrained network to featurize data for cats vs dogs

featurize.py

from keras.applications.resnet50 import ResNet50, preprocess_input
from keras.preprocessing.image import load_img, img_to_array
import numpy as np
from os import listdir
from os.path import join
import pandas as pd


def featurize(fpath, model, img_size):
    '''Use trained model to convert image to vector describing content

    Arguments
    ---------
        fpath: string with path to file to featurize
        model: Keras model object used to featurize image
        img_size: Tuple with size image should be rescaled to during file loading. Must match model's input layer
    '''
    
    img = load_img(fpath, target_size=img_size)
    raw_arr = img_to_array(img)
    raw_arr = np.expand_dims(raw_arr, axis=0)
    raw_arr = preprocess_input(raw_arr)
    features = model.predict(raw_arr).flatten()
    return features

if __name__ == "__main__":
    data_dir = 'raw'
    fnames = listdir(data_dir)
    fpaths = (join(data_dir, f) for f in fnames)

    model = ResNet50(weights='imagenet', include_top=False)
    img_size = (224, 224)  # image size expected by ResNet model
    x = np.vstack([featurize(fpath, model, img_size) for fpath in image_paths])
    y = np.array(['dog' in fname for fname in fnames])