Application of AutoKeras for Deep Learning: Notebook in Python for Beginners to Professionals

AutoKeras Project – A Guide to build a wine quality classification model in Python

AutoKeras for Beginners - A Guide to build a wine quality classification model in Python

For more projects visit: https://setscholars.net

• There are 5000+ free end-to-end applied machine learning and data science projects available to download at SETSscholar. SETScholars is a Science, Engineering and Technology Scholars community.

# Suppress warnings in Jupyter Notebooks
import warnings
warnings.filterwarnings("ignore")

import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
import tensorflow as tf
import tensorflow_datasets as tfds
import autokeras as ak

print(ak.__version__)

import logging
tf.get_logger().setLevel(logging.ERROR)

1.0.16

In this notebook, we will learn how to build a wine quality classification model in Python using AutoKeras package.

Python Codes

Load Dataset from Tensorflow built-in datasets

ds, info = tfds.load('wine_quality', split = 'train', shuffle_files=True, with_info=True)
df = tfds.as_dataframe(ds)

print(); print(df.shape)
print(); print(df.head())
print(); print(df.columns.values)

(4898, 12)

   features/alcohol  features/chlorides  features/citric acid  \
0               9.0               0.054                  0.34   
1              12.2               0.063                  0.49   
2              11.2               0.029                  0.11   
3              10.3               0.055                  0.39   
4              10.7               0.054                  0.35   

   features/density  features/fixed acidity  features/free sulfur dioxide  \
0           1.00080                     7.6                          44.0   
1           0.99110                     6.3                          35.0   
2           0.99076                     5.3                           6.0   
3           0.99652                     7.0                          42.0   
4           0.99178                     7.3                          31.0   

   features/pH  features/residual sugar  features/sulphates  \
0         3.22                    18.35                0.55   
1         3.38                     1.20                0.42   
2         3.51                     1.10                0.48   
3         3.37                     7.50                0.54   
4         3.18                     1.60                0.47   

   features/total sulfur dioxide  features/volatile acidity  quality  
0                          197.0                       0.32        5  
1                           92.0                       0.27        6  
2                           51.0                       0.43        4  
3                          218.0                       0.31        5  
4                          148.0                       0.28        5  

['features/alcohol' 'features/chlorides' 'features/citric acid'
 'features/density' 'features/fixed acidity'
 'features/free sulfur dioxide' 'features/pH' 'features/residual sugar'
 'features/sulphates' 'features/total sulfur dioxide'
 'features/volatile acidity' 'quality']

Profiling dataset and EDA with visualisation

#import pandas_profiling
#df.profile_report()

#import sweetviz as sv
#sweet_report = sv.analyze(df)
#sweet_report.show_notebook(layout='vertical', w=880, h=1000,scale=0.8)

Train and Test data split

y = df.pop('quality')
X = df

print(y.shape)
print(X.shape)

(4898,)
(4898, 11)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

print(); print("Training Dataset:")
print(); print(X_train.shape)
print(); print(X_train.head())
print(); print(y_train.shape)
print(); print(y_train.head())

print(); print("\n\nTesting Dataset:")
print(); print(X_test.shape)
print(); print(X_test.head())
print(); print(y_test.shape)
print(); print(y_test.head())

Training Dataset:

(3281, 11)

      features/alcohol  features/chlorides  features/citric acid  \
3167             12.50               0.030                  0.43   
324               9.75               0.069                  0.47   
4731              9.10               0.050                  0.30   
2305              9.50               0.039                  0.47   
1777             12.10               0.036                  0.28   

      features/density  features/fixed acidity  features/free sulfur dioxide  \
3167           0.99164                     6.8                          30.0   
324            0.99391                     5.7                          35.0   
4731           0.99652                     7.2                          40.0   
2305           0.99590                     9.5                          21.0   
1777           0.99206                     6.6                          10.0   

      features/pH  features/residual sugar  features/sulphates  \
3167         3.08                      7.6                0.59   
324          3.11                      6.3                0.46   
4731         3.15                      8.1                0.49   
2305         2.90                      1.3                0.64   
1777         3.07                      9.2                0.35   

      features/total sulfur dioxide  features/volatile acidity  
3167                          110.0                       0.28  
324                           182.0                       0.24  
4731                          188.0                       0.30  
2305                          123.0                       0.21  
1777                           92.0                       0.39  

(3281,)

3167    8
324     5
4731    6
2305    5
1777    6
Name: quality, dtype: int64



Testing Dataset:

(1617, 11)

      features/alcohol  features/chlorides  features/citric acid  \
4656              10.7               0.048                  0.28   
3659               8.7               0.122                  0.99   
907                9.4               0.042                  0.72   
4352               9.5               0.054                  0.26   
3271              13.0               0.021                  0.29   

      features/density  features/fixed acidity  features/free sulfur dioxide  \
4656           0.99556                     6.8                          54.0   
3659           0.99360                     6.6                          45.0   
907            0.99990                     8.0                          62.0   
4352           0.99538                     6.4                          47.0   
3271           0.99026                     7.5                          38.0   

      features/pH  features/residual sugar  features/sulphates  \
4656         3.19                12.600000                0.37   
3659         3.09                 1.200000                0.31   
907          2.92                17.549999                0.68   
4352         3.12                 8.200000                0.50   
3271         3.08                 4.900000                0.48   

      features/total sulfur dioxide  features/volatile acidity  
4656                          136.0                       0.20  
3659                          129.0                       0.19  
907                           233.0                       0.66  
4352                          182.0                       0.24  
3271                          113.0                       0.38  

(1617,)

4656    6
3659    6
907     4
4352    5
3271    7
Name: quality, dtype: int64

print(y_train.dtype)
print(y_test.dtype)

int64
int64

y_train = y_train.astype('str')
y_test  = y_test.astype('str')

print(y_train.dtype)
print(y_test.dtype)

object
object

Training of AutoKeras Model for Structre Dataset (Tabular dataset)

# It tries 10 different models.
clf = ak.StructuredDataClassifier(overwrite=True, max_trials=10)

# Feed the structured data classifier with training data.
clf.fit(X_train, y_train, validation_split=0.20, epochs=50, batch_size=32, verbose=1)

Trial 10 Complete [00h 00m 13s]
val_accuracy: 0.553600013256073

Best val_accuracy So Far: 0.553600013256073
Total elapsed time: 00h 01m 37s
Epoch 1/50
103/103 [==============================] - 1s 1ms/step - loss: 1.8509 - accuracy: 0.2509
Epoch 2/50
103/103 [==============================] - 0s 1ms/step - loss: 1.2746 - accuracy: 0.4719
Epoch 3/50
103/103 [==============================] - 0s 1ms/step - loss: 1.2128 - accuracy: 0.4871
Epoch 4/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1861 - accuracy: 0.4700
Epoch 5/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1632 - accuracy: 0.4973
Epoch 6/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1536 - accuracy: 0.5011
Epoch 7/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1356 - accuracy: 0.4951
Epoch 8/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1318 - accuracy: 0.4950
Epoch 9/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1215 - accuracy: 0.5002
Epoch 10/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1160 - accuracy: 0.5070
Epoch 11/50
103/103 [==============================] - 0s 1ms/step - loss: 1.1119 - accuracy: 0.5140
Epoch 12/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0918 - accuracy: 0.5084
Epoch 13/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0958 - accuracy: 0.5184
Epoch 14/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0868 - accuracy: 0.5143
Epoch 15/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0791 - accuracy: 0.5249
Epoch 16/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0847 - accuracy: 0.5323
Epoch 17/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0681 - accuracy: 0.5290
Epoch 18/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0656 - accuracy: 0.5385
Epoch 19/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0605 - accuracy: 0.5324
Epoch 20/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0601 - accuracy: 0.5245
Epoch 21/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0580 - accuracy: 0.5306
Epoch 22/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0575 - accuracy: 0.5351
Epoch 23/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0496 - accuracy: 0.5429
Epoch 24/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0412 - accuracy: 0.5297
Epoch 25/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0386 - accuracy: 0.5517
Epoch 26/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0361 - accuracy: 0.5411
Epoch 27/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0350 - accuracy: 0.5617
Epoch 28/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0331 - accuracy: 0.5599
Epoch 29/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0289 - accuracy: 0.5561
Epoch 30/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0296 - accuracy: 0.5480
Epoch 31/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0254 - accuracy: 0.5473
Epoch 32/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0232 - accuracy: 0.5487
Epoch 33/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0121 - accuracy: 0.5571
Epoch 34/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0188 - accuracy: 0.5425
Epoch 35/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0132 - accuracy: 0.5652
Epoch 36/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0169 - accuracy: 0.5537
Epoch 37/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0123 - accuracy: 0.5548
Epoch 38/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0189 - accuracy: 0.5583
Epoch 39/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0151 - accuracy: 0.5540
Epoch 40/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0102 - accuracy: 0.5642
Epoch 41/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9941 - accuracy: 0.5637
Epoch 42/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9983 - accuracy: 0.5735
Epoch 43/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0053 - accuracy: 0.5624
Epoch 44/50
103/103 [==============================] - 0s 1ms/step - loss: 1.0060 - accuracy: 0.5546
Epoch 45/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9978 - accuracy: 0.5671
Epoch 46/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9907 - accuracy: 0.5810
Epoch 47/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9902 - accuracy: 0.5669
Epoch 48/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9789 - accuracy: 0.5705
Epoch 49/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9830 - accuracy: 0.5704
Epoch 50/50
103/103 [==============================] - 0s 1ms/step - loss: 0.9828 - accuracy: 0.5825

<tensorflow.python.keras.callbacks.History at 0x7f007ff510d0>

Performance of the AutoKeras Model

# Evaluate the best model with testing data.
print(); print();
print(clf.evaluate(X_test, y_test, verbose=0))

# Evaluate the model on the test data using `evaluate`
print()
print("Evaluate on test data")
results = clf.evaluate(X_test, y_test, batch_size=32, verbose=0)
print("test loss,     test acc: \n", results)

[1.1242220401763916, 0.5275201201438904]

Evaluate on test data
test loss,     test acc: 
 [1.1242220401763916, 0.5275201201438904]

import scikitplot as skplt
from sklearn.metrics import accuracy_score, classification_report
from sklearn.metrics import cohen_kappa_score, confusion_matrix

# Predict with the best model.
predicted_y = clf.predict(X_test, verbose=0)

# Evaluate the skill of the Trained model
acc                 = accuracy_score(y_test, predicted_y)
classReport         = classification_report(y_test, predicted_y)
confMatrix          = confusion_matrix(y_test, predicted_y) 
        
print(); print('Testing Results of the trained model: ')
print(); print('Accuracy : ', acc)
print(); print('Confusion Matrix :\n', confMatrix)
print(); print('Classification Report :\n',classReport)

# Confusion matrix
skplt.metrics.plot_confusion_matrix(y_test, predicted_y,figsize=(7,7)); plt.show()

Testing Results of the trained model: 

Accuracy :  0.5275200989486704

Confusion Matrix :
 [[  0   2   2   3   0   0   0]
 [  0   1  36  12   1   0   0]
 [  0   4 282 200   4   0   0]
 [  0   2 179 487  38   0   0]
 [  0   0  23 180  83   0   0]
 [  0   0   7  59   9   0   0]
 [  0   0   0   1   2   0   0]]

Classification Report :
               precision    recall  f1-score   support

           3       0.00      0.00      0.00         7
           4       0.11      0.02      0.03        50
           5       0.53      0.58      0.55       490
           6       0.52      0.69      0.59       706
           7       0.61      0.29      0.39       286
           8       0.00      0.00      0.00        75
           9       0.00      0.00      0.00         3

    accuracy                           0.53      1617
   macro avg       0.25      0.23      0.22      1617
weighted avg       0.50      0.53      0.50      1617

Model Architecture

model = clf.export_model()

model.summary()

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 11)]              0         
_________________________________________________________________
multi_category_encoding (Mul (None, 11)                0         
_________________________________________________________________
normalization (Normalization (None, 11)                23        
_________________________________________________________________
dense (Dense)                (None, 32)                384       
_________________________________________________________________
re_lu (ReLU)                 (None, 32)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 32)                1056      
_________________________________________________________________
re_lu_1 (ReLU)               (None, 32)                0         
_________________________________________________________________
dropout (Dropout)            (None, 32)                0         
_________________________________________________________________
dense_2 (Dense)              (None, 7)                 231       
_________________________________________________________________
classification_head_1 (Softm (None, 7)                 0         
=================================================================
Total params: 1,694
Trainable params: 1,671
Non-trainable params: 23
_________________________________________________________________

from tensorflow.keras.utils import plot_model
plot_model(model)

Save model

print(type(model))  

try:
    model.save("best_keras_model", save_format="tf")
except Exception:
    model.save("best_keras_model.h5")

<class 'tensorflow.python.keras.engine.functional.Functional'>

Load model

model = tf.keras.models.load_model('best_keras_model', custom_objects=ak.CUSTOM_OBJECTS)

Prediction using the model

# Predict with the best model
predicted_probablity = model.predict(X_test, verbose=1)
print(); print(predicted_probablity)

51/51 [==============================] - 0s 689us/step

[[3.1682735e-04 5.8419011e-03 1.6993020e-01 ... 8.6702302e-02
  3.6141742e-02 3.1056977e-04]
 [1.1442063e-03 2.5611144e-02 2.4990357e-01 ... 2.0381109e-01
  2.6276793e-02 8.9784770e-04]
 [2.7582890e-04 4.5949418e-02 9.1155684e-01 ... 6.2132633e-04
  2.9407036e-06 1.0756439e-09]
 ...
 [4.9888232e-04 7.0865517e-03 1.3117082e-01 ... 3.2787517e-01
  1.4668092e-02 1.7885838e-04]
 [5.1948999e-04 4.7347620e-02 5.7962453e-01 ... 4.4063956e-02
  1.6590484e-03 7.5194264e-05]
 [5.1107685e-07 1.2151637e-03 4.4980764e-01 ... 1.3518638e-02
  1.0267833e-04 5.0248076e-08]]

Summary

In this coding recipe, we discussed how to build a Classification Model in Python using AutoKeras.

Specifically, we have learned the followings:

• How to build a wine quality classification model in Python using AutoKeas.
• How to load a tensorflow built-in dataset for machine learning practice.