demo-mini.py

#!/usr/bin/env python
# coding: utf-8

# ## Fruit Image Classification
# ### data source: https://www.kaggle.com/moltean/fruits

# In[1]:


import time
start = time.time()


# In[2]:


import warnings
warnings.filterwarnings("ignore")
import os
#import pandas as pd
import numpy as np
import tensorflow 
import keras
import h5py


# ### multiclass classification 
# #### four fruits selected for mini dataset

# In[3]:


from sklearn.datasets import load_files


# In[4]:


def load_dataset(path):
    data=load_files(path)
    files = np.array(data['filenames'])
    targets = np.array(data['target'])
    target_labels = np.array(data['target_names'])
    return files, targets, target_labels


# In[5]:


X_train, y_train, target_labels = load_dataset('fruits-360-cp/Training_mini')


# In[6]:


print('training data loaded')


# In[7]:


X_test, y_test, _ = load_dataset('fruits-360-cp/Test_mini')


# In[8]:


print('test data loaded')


# In[9]:


print('Training set:', X_train.shape[0])
print('Test set:', X_test.shape[0])


# In[10]:


n_class=len(np.unique(y_train))


# In[11]:


print('number of categories: ', n_class)


# In[12]:


from keras.utils import np_utils
y_train=np_utils.to_categorical(y_train, n_class)
y_test=np_utils.to_categorical(y_test, n_class)


# In[13]:


from keras.preprocessing.image import array_to_img, img_to_array, load_img


# In[14]:


def convert_to_array(pics):
    img_arr =[]
    for pic in pics:
        img_arr.append(img_to_array(load_img(pic)))
    return img_arr


# In[15]:


X_train = np.array(convert_to_array(X_train))
X_test = np.array(convert_to_array(X_test))


# In[16]:


X_train = X_train.astype('float32')/255
X_test = X_test.astype('float32')/255


# In[17]:


print('data reshaped')
print('sample shape for model input: ',X_train[0].shape)


# In[18]:


#import all the pieces of the neural network
from keras.models import Sequential, Model#the model
from keras.layers import Dense, Dropout, Activation, Flatten, Input #fully connected layers
from keras.layers import Convolution2D, MaxPooling2D #the convnet
from keras.utils import np_utils #extra tools
from keras.callbacks import ModelCheckpoint


# In[19]:


# filepath="weights_best"
# checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')
# callbacks_list = [checkpoint]


# In[20]:


# from keras.callbacks import History #may want to print history if set verbose to 0
# history = History()


# In[21]:


# load the weights that yielded the best validation accuracy
#model.load_weights('cnn_test.hdf5')


# In[22]:


print('running Keras CNN model')


# In[23]:


#THE MODEL#
batch_size = 32
nb_classes = n_class
nb_epoch = 2

img_rows =100
img_cols=100
img_channels = 3

model_input=Input(shape=(img_rows, img_cols,img_channels))

x = Convolution2D(32, 3, 3, border_mode='same')(model_input)
x = Activation('relu')(x)
x = Convolution2D(32, 3, 3)(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Dropout(0.25)(x)
conv_out = Flatten()(x)

x1 = Dense(nb_classes, activation='softmax')(conv_out)

lst = [x1]

#model = Model(input=model_input, output=lst)
model = Model(input=model_input, output=lst) 

model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])

model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=nb_epoch, verbose=1)
#set verbose to 0 if you run many, 


# In[24]:


y_pred = model.predict(X_test)


# In[25]:


score=model.evaluate(X_test, y_test, verbose=0)
print('Test accuracy:', score[1])


# In[26]:


end = time.time()
print("total runtime: ",(end - start))


# In[27]:


import matplotlib.pyplot as plt
#get_ipython().run_line_magic('matplotlib', 'inline')
# plot a random sample of test images, their predicted labels, and ground truth
fig = plt.figure(figsize=(16, 9))
for i, idx in enumerate(np.random.choice(X_test.shape[0], size=16, replace=False)):
    ax = fig.add_subplot(4, 4, i + 1, xticks=[], yticks=[])
    ax.imshow(np.squeeze(X_test[idx]))
    pred_idx = np.argmax(y_pred[idx])
    true_idx = np.argmax(y_test[idx])
    ax.set_title("{} ({})".format(target_labels[pred_idx], target_labels[true_idx]),
                 color=("green" if pred_idx == true_idx else "red"))

plt.savefig('demo-mini.png')

from PIL import Image
img=Image.open('demo-mini.png')
img.show()