import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import keras
from IPython.display import Image
import cv2

import warnings
warnings.filterwarnings("ignore")

from keras.models import Model
from keras.utils import plot_model,to_categorical
from keras.layers import Conv2D,Dense,Flatten,BatchNormalization,Add,Activation,Dropout,\
            MaxPooling2D,AveragePooling2D,concatenate
from keras.engine.input_layer import Input
from keras.initializers import glorot_uniform
from keras.regularizers import l2
from keras.optimizers import Adam


from keras.datasets import cifar10

(x_train,y_train),(x_test,y_test) = cifar10.load_data()

x_train = x_train[:10000]
y_train = y_train[:10000]

x_train = np.array([cv2.resize(img, (224,224)) for img in x_train[:,:,:,:]])
x_test = np.array([cv2.resize(img, (224,224)) for img in x_test[:,:,:,:]])

y_train = to_categorical(y_train)

x_train = x_train.astype('float32')/ 255
x_test = x_test.astype('float32')/255


def inception_layer(inputs,filter_1,filter_2_1,filter_2_2,
             filter_3_1,filter_3_2,filter_4):
#     print(inputs.shape)

    Conv1x1 = Conv2D(filter_1,kernel_size=(1,1),strides=1,kernel_initializer = glorot_uniform(),
                     activation='relu',padding='SAME')(inputs)

    Conv3x3 = Conv2D(filter_2_1,kernel_size=(1,1),strides=1,kernel_initializer = glorot_uniform(),
                       activation='relu',padding='SAME')(inputs)
#     print(Conv3x3.shape)
    Conv3x3 = Conv2D(filter_2_2,kernel_size=(3,3),strides=1,kernel_initializer = glorot_uniform(),
                       activation='relu',padding='SAME')(Conv3x3)
#     print(Conv3x3.shape)
    Conv5x5 = Conv2D(filter_3_1,kernel_size=(1,1),strides=1,kernel_initializer= glorot_uniform(),
                       activation='relu',padding='SAME')(inputs)
#     print(Conv5x5.shape)
    Conv5x5 = Conv2D(filter_3_2,kernel_size=(5,5),strides=1,kernel_initializer=glorot_uniform(),
                       activation='relu',padding='SAME')(Conv5x5)
#     print(Conv5x5.shape)

    pool_proj = MaxPooling2D((3,3),strides=1,padding='SAME')(inputs)
    pool_proj = Conv2D(filter_4,kernel_size=(1,1),strides=1,kernel_initializer = glorot_uniform(),
                       activation='relu',padding='SAME')(pool_proj)

    output = concatenate([Conv1x1,Conv3x3,Conv5x5,pool_proj],axis=3)

    return output


input_shape = x_train.shape[1:]
input_layer = Input(shape=input_shape)

x = Conv2D(64,kernel_size=(7,7),strides=(2,2),padding='SAME',activation='relu',
               kernel_initializer = glorot_uniform())(input_layer)
x = MaxPooling2D((3,3),2,padding='SAME')(x)
x = Conv2D(64,kernel_size=(3,3),strides=1,padding='SAME',activation='relu',
               kernel_initializer = glorot_uniform())(x)
x = Conv2D(192,kernel_size=(3,3),strides=1,padding='SAME',activation='relu',
               kernel_initializer = glorot_uniform())(x)
x = MaxPooling2D((3,3),2,padding='SAME')(x)

inception3a = inception_layer(x,64,96,128,16,32,32)
inception3b = inception_layer(inception3a,128,128,192,32,96,64)

x = MaxPooling2D((3,3),2,padding='SAME')(inception3b)

inception4a = inception_layer(x,192,96,208,16,48,64)
inception4b = inception_layer(inception4a,160,112,224,24,64,64)
inception4c = inception_layer(inception4b,128,128,256,24,64,64)
inception4d = inception_layer(inception4c,112,114,288,32,64,64)
inception4e = inception_layer(inception4d,256,160,320,32,128,128)

x = MaxPooling2D((3,3),3,padding='SAME')(inception4e)

inception5a = inception_layer(x,256,160,320,32,128,128)
inception5b = inception_layer(inception5a,384,192,384,48,128,128)

x = AveragePooling2D((4,4),1)(inception5b)
x = Dropout(0.40)(x)

x  = Flatten()(x)
x = Dense(10)(x)
x = Activation('softmax')(x)

model = Model(input_layer,x)
model.compile(loss='categorical_crossentropy',optimizer='sgd',metrics=['accuracy'])

history = model.fit(x_train,y_train,validation_split=0.1,epochs=5,batch_size=128)