📑 Neural Networks for Classification of Drawings & Photo Images

🌐   DL_PP8_0_SMC.html

📑 Deep Learning. Practice Project 8_0:
Neural Networks for Classification of Drawings & Photo Images

✒️ Code Modules, Helpful Functions, & Settings

xxxxxxxxxx

 

1

#!python3 -m pip install tensorflow==2.6.0 \

2

#--user --quiet --no-warn-script-location

3

path='/home/sc_work/.sage/local/lib/python3.9/site-packages'

4

import sys,warnings; sys.path.append(path)

5

warnings.filterwarnings('ignore')

6

import os,h5py,urllib,pandas as pd,numpy as np

7

import tensorflow as tf,pylab as pl,seaborn as sn

8

import tensorflow.keras.layers as tkl

9

import tensorflow.keras.callbacks as tkc

10

from sklearn.metrics import classification_report

11

file_path='https://raw.githubusercontent.com/'+\

12

'OlgaBelitskaya/data_kitchen/main/'

13

file_name='Pictograms64.h5'

14

model_weights='/tmp/checkpoint'

15

img_size=int(32)

16

from IPython.display import display,HTML

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

def h5file2data(h5file,cmap,img_size,resize=True):

2

    with h5py.File(h5file,'r') as f:

3

        keys=list(f.keys())

4

        pretty_print('file keys: '+', '.join(keys))

5

        images=np.array(f[keys[int(0)]])

6

        labels=np.array(f[keys[int(1)]])

7

        names=[[el.decode('utf-8') for el in f[keys[i]]]

8

               for i in range(int(2),len(keys))]

9

        f.close()

10

    N=images.shape[int(0)]; n=int(.1*N)

11

    if resize:

12

        images=tf.image.resize(images,[img_size,img_size]).numpy()

13

    shuffle_ids=np.arange(N)

14

    np.random.RandomState(12).shuffle(shuffle_ids)

15

    images=images[shuffle_ids]

16

    labels=np.array([labels[i][shuffle_ids] 

17

                     for i in range(labels.shape[int(0)])])

18

    x_test,x_valid,x_train=images[:n],images[n:2*n],images[2*n:]

19

    y_test,y_valid,y_train=\

20

    labels[:,:n],labels[:,n:2*n],labels[:,2*n:]

21

    pretty_print('data outputs: ')

22

    df=pd.DataFrame([[x_train.shape,x_valid.shape,x_test.shape],

23

                     [x_train.dtype,x_valid.dtype,x_test.dtype],

24

                     [y_train.shape,y_valid.shape,y_test.shape],

25

                     [y_train.dtype,y_valid.dtype,y_test.dtype]],

26

                    columns=['train','valid','test'],

27

                    index=['image shape','image type',

28

                           'label shape','label type'])

29

    display(df)

30

    pretty_print('distribution of labels: ')

31

    idx=['labels %d'%(i+int(1)) for i in range(labels.shape[int(0)])]

32

    df=pd.DataFrame(labels,index=idx).T

33

    for i in range(labels.shape[int(0)]):

34

        df['name %d'%(i+int(1))]=[names[i][l] for l in labels[i]]

35

    fig=pl.figure(figsize=(6,6))    

36

    for i in range(labels.shape[int(0)]):

37

        ax=fig.add_subplot(labels.shape[int(0)],int(1),i+int(1))

38

        sn.countplot(y='name %s'%(i+1),data=df,

39

                     palette=cmap,alpha=.5,ax=ax)

40

    pl.tight_layout(); pl.show()       

41

    return [names,x_train,x_valid,x_test,y_train,y_valid,y_test]

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

def display_images(images,labels,names,n):

2

    fig=pl.figure(figsize=(6,n/2))

3

    randch=np.random.choice(images.shape[0],size=n,replace=False)

4

    for i,idx in enumerate(randch):

5

        ax=fig.add_subplot(int(n//3)+1,3,i+1,xticks=[],yticks=[])

6

        ax.imshow(images[idx],cmap='bone')

7

        if len(labels.shape)<int(2): 

8

            label=[labels[idx]]

9

            name=[names[labels[idx]]]

10

        else: 

11

            label=[labels[:,idx]]

12

            name=[names[i][labels[i][idx]] 

13

                  for i in range(labels.shape[0])]

14

        ti='{} \n {}'.format(str(label),str(name))

15

        ax.set_title(ti,fontsize=8)

16

    pl.tight_layout(); pl.show()

17

def keras_history_plot(fit_history,fig_size=6,color='#348ABD'):

18

    keys=list(fit_history.history.keys())

19

    list_history=[fit_history.history[keys[i]] 

20

                  for i in range(len(keys))]

21

    dfkeys=pd.DataFrame(list_history).T

22

    dfkeys.columns=keys

23

    fig=pl.figure(figsize=(fig_size,fig_size//2))

24

    ax1=fig.add_subplot(2,1,1)

25

    dfkeys.iloc[:,[int(0),int(2)]].plot(

26

        ax=ax1,color=['slategray',color],grid=True)

27

    ax2=fig.add_subplot(2,1,2)

28

    dfkeys.iloc[:,[int(1),int(3)]].plot(

29

        ax=ax2,color=['slategray',color],grid=True)

30

    pl.tight_layout(); pl.show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️ Pictogram's Data Loading

xxxxxxxxxx

 

1

input_file=urllib.request.urlopen(file_path+file_name)

2

output_file=open(file_name,'wb')

3

output_file.write(input_file.read())

4

output_file.close(); input_file.close()

5

[names,x_train,x_valid,x_test,y_train,y_valid,y_test]=\

6

h5file2data(file_name,'Pastel1',img_size)

7

os.remove(file_name)

8

xd=np.vstack([x_train,x_valid,x_test])

9

yd=np.hstack([y_train,y_valid,y_test])

10

display_images(xd,yd,names,6)

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️ Pictogram Selecting

xxxxxxxxxx

 

1

names2=['plane','car','bird','cat','deer',

2

        'dog','frog','horse','ship','truck']

3

xd=xd[np.where(yd[0]==0)]

4

yd=yd[1][np.where(yd[0]==0)]

5

cond=np.where([l in names2 for l in names[1]])[0]

6

cond2=np.where([l in cond for l in yd])

7

xd=xd[cond2]; yd=yd[cond2]

8

rd={1:2,4:0,6:5,7:7,8:4,9:9,10:1,11:3,12:6,13:8}

9

yd=np.array([rd.get(x,x) for x in yd],dtype='int8')

10

display_images(xd,yd,names2,6)

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️ CIFAR's Data Loading

xxxxxxxxxx

 

1

file_name='CIFAR10_5000.h5'

2

input_file=urllib.request.urlopen(file_path+file_name)

3

output_file=open(file_name,'wb')

4

output_file.write(input_file.read())

5

output_file.close(); input_file.close()

6

with h5py.File(file_name,'r') as f:

7

    keys=list(f.keys())

8

    pretty_print('file keys: '+', '.join(keys))

9

    x=np.array(f[keys[int(0)]])

10

    y=np.array(f[keys[int(1)]])

11

    f.close()

12

n2=int(1.2*len(yd)); N=len(y); n=int(.1*N)

13

shuffle_ids=np.arange(N)

14

np.random.RandomState(23).shuffle(shuffle_ids)

15

shuffle_ids=shuffle_ids[:n2]

16

x,y=x[shuffle_ids],y[shuffle_ids]

17

display_images(x,y,names2,6)

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️ Mixed Data

xxxxxxxxxx

 

1

x=np.vstack([x,xd]); y=np.hstack([y,yd])

2

N=len(y); n=int(.1*N)

3

shuffle_ids=np.arange(N)

4

np.random.RandomState(23).shuffle(shuffle_ids)

5

x,y=x[shuffle_ids],y[shuffle_ids]

6

print([x.shape,x.dtype,y.shape,y.dtype])

7

gx=np.dot(x[...,:3],[.299,.587,.114])

8

x_test,x_valid,x_train=x[:n],x[n:2*n],x[2*n:]

9

gx_test,gx_valid,gx_train=gx[:n],gx[n:2*n],gx[2*n:]

10

y_test,y_valid,y_train=y[:n],y[n:2*n],y[2*n:]

11

display_images(x_train,y_train,names2,3)

12

display_images(gx_train,y_train,names2,3)

13

gx_train=gx_train.reshape(-int(1),img_size,img_size,int(1))

14

gx_valid=gx_valid.reshape(-int(1),img_size,img_size,int(1))

15

gx_test=gx_test.reshape(-int(1),img_size,img_size,int(1))

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

df=pd.DataFrame(y,columns=['label'])

2

df['class']=[names2[l] for l in y]

3

pl.figure(figsize=(6,4))

4

sn.countplot(y='class',data=df,palette='Pastel1',alpha=.5)

5

pl.title('distribution of labels',fontsize=15)

6

pl.tight_layout(); pl.show()

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

✒️ NN Classification of Colored Images

xxxxxxxxxx

 

1

def model(leaky_alpha):

2

    model=tf.keras.Sequential()

3

    model.add(tkl.Conv2D(

4

        int(32),(int(5),int(5)),padding='same', 

5

        input_shape=x_train.shape[int(1):]))

6

    model.add(tkl.LeakyReLU(alpha=leaky_alpha))    

7

    model.add(tkl.MaxPooling2D(pool_size=(int(2),int(2))))

8

    model.add(tkl.Dropout(.2))

9

    model.add(tkl.Conv2D(int(96),(int(5),int(5))))

10

    model.add(tkl.LeakyReLU(alpha=leaky_alpha))    

11

    model.add(tkl.MaxPooling2D(pool_size=(int(2),int(2))))

12

    model.add(tkl.Dropout(.2))   

13

    model.add(tkl.GlobalMaxPooling2D())     

14

    model.add(tkl.Dense(int(512)))

15

    model.add(tkl.LeakyReLU(alpha=leaky_alpha))

16

    model.add(tkl.Dropout(.5))     

17

    model.add(tkl.Dense(int(10)))

18

    model.add(tkl.Activation('softmax'))   

19

    model.compile(loss='sparse_categorical_crossentropy',

20

                  optimizer='nadam',metrics=['accuracy'])   

21

    return model

22

model=model(float(.005))

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

checkpointer=tkc.ModelCheckpoint(

2

    filepath=model_weights,verbose=int(0),save_weights_only=True,

3

    monitor='val_accuracy',mode='max',save_best_only=True)

4

lr_reduction=tkc.ReduceLROnPlateau(

5

    monitor='val_loss',patience=int(5),verbose=int(0),factor=.8)

6

history=model.fit(x_train,y_train,epochs=int(8),

7

                  batch_size=int(16),verbose=int(2),

8

                  validation_data=(x_valid,y_valid),

9

                  callbacks=[checkpointer,lr_reduction])

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages

xxxxxxxxxx

 

1

keras_history_plot(history)

2

model.load_weights(model_weights)

3

pretty_print(model.evaluate(x_test,y_test,verbose=int(0)))

4

py_test=np.argmax(model.predict(x_test),axis=-1)

5

print(classification_report(y_test,py_test))

Language: Sage Gap GP HTML Macaulay2 Maxima Octave Python R Singular

---

Messages