fashion.py

# https://deeplearningcourses.com/c/advanced-computer-vision
# https://www.udemy.com/advanced-computer-vision

from __future__ import print_function, division
from builtins import range
# Note: you may need to update your version of future
# sudo pip install -U future

from keras.models import Sequential
from keras.layers import Dense, Activation, Conv2D, MaxPooling2D, Flatten, Dropout, BatchNormalization

import matplotlib.pyplot as plt
import pandas as pd
import numpy as np


# helper
def y2indicator(Y):
  N = len(Y)
  K = len(set(Y))
  I = np.zeros((N, K))
  I[np.arange(N), Y] = 1
  return I


# get the data
# https://www.kaggle.com/zalando-research/fashionmnist
data = pd.read_csv('../large_files/fashionmnist/fashion-mnist_train.csv')
data = data.values
np.random.shuffle(data)

X = data[:, 1:].reshape(-1, 28, 28, 1) / 255.0
Y = data[:, 0].astype(np.int32)

# get shapes
# N = len(Y)
K = len(set(Y))

# by default Keras wants one-hot encoded labels
# there's another cost function we can use
# where we can just pass in the integer labels directly
# just like Tensorflow / Theano
Y = y2indicator(Y)


# the model will be a sequence of layers
model = Sequential()


# make the CNN
# model.add(Input(shape=(28, 28, 1)))
model.add(Conv2D(input_shape=(28, 28, 1), filters=32, kernel_size=(3, 3)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())

model.add(Conv2D(filters=64, kernel_size=(3, 3)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())

model.add(Conv2D(filters=128, kernel_size=(3, 3)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D())

model.add(Flatten())
model.add(Dense(units=300))
model.add(Activation('relu'))
model.add(Dropout(0.2))
model.add(Dense(units=K))
model.add(Activation('softmax'))


# list of losses: https://keras.io/losses/
# list of optimizers: https://keras.io/optimizers/
# list of metrics: https://keras.io/metrics/
model.compile(
  loss='categorical_crossentropy',
  optimizer='adam',
  metrics=['accuracy']
)

# note: multiple ways to choose a backend
# either theano, tensorflow, or cntk
# https://keras.io/backend/


# gives us back a <keras.callbacks.History object at 0x112e61a90>
r = model.fit(X, Y, validation_split=0.33, epochs=15, batch_size=32)
print("Returned:", r)

# print the available keys
# should see: dict_keys(['val_loss', 'acc', 'loss', 'val_acc'])
print(r.history.keys())

# plot some data
plt.plot(r.history['loss'], label='loss')
plt.plot(r.history['val_loss'], label='val_loss')
plt.legend()
plt.show()

# accuracies
plt.plot(r.history['accuracy'], label='acc')
plt.plot(r.history['val_accuracy'], label='val_acc')
plt.legend()
plt.show()