timehaven

cd /path/to/demo/data/

# Download train.zip from Kaggle using
# https://www.kaggle.com/c/dogs-vs-cats/data

ls -l train.zip

-rw-r--r--@ 1 user  group   543M Jun 12 10:39 train.zip

unzip -qq train.zip

timehaven

cd /path/to/kaggle/data/
ls train/dogs| head

dog.1000.jpg
dog.1001.jpg
dog.1002.jpg
dog.1003.jpg
dog.1004.jpg
dog.1005.jpg
dog.1006.jpg

timehaven

cd data
rm train validation  # symbolic links to _demo dirs.

mkdir -p train_orig/dogs train_orig/cats \
  validation_orig/dogs validation_orig/cats

mv train_orig/dog*[12][345]*.jpg validation_orig/dogs
mv train_orig/cat*[12][345]*.jpg validation_orig/cats
mv train_orig/dog*.jpg train_orig/dogs
mv train_orig/cat*.jpg train_orig/cats

timehaven

# 'features' is file path to bcolz array on disk
bc = bcolz.open(features)[:]

# begin epoch loop
while 1:
    ...
    df = df.sample(frac=1)  # shuffle all rows
    bc = bc[df.index.values]
    ...

timehaven

...
    i, j = 0, batch_size
    for _ in range(nbatches):
        sub = df.iloc[i:j]
        X2 = bc[i:j]
        ...
        # Calculate X and Y appropriately
        ...
        yield [X, X2], Y
        i = j

timehaven

df_mini_batch

13      95555756  dog   grey  /path/to/imgs/756/55/blah_95555756.png
5        5467756  cat  black  /path/to/imgs/756/67/blah_5467756.png
1        1161756  cat  black  /path/to/imgs/756/61/blah_1161756.png
7       31255756  cat   grey  /path/to/imgs/756/55/blah_31255756.png

timehaven

while 1:
    ...
    df = df.sample(frac=1)  # shuffle all rows
    ...
    i, j = 0, batch_size
    for _ in range(nbatches):
        sub = df.iloc[i:j]
        idx = sub.index.values
        X2 = bcolz.open(bcolz_dir)[idx]
        ...