Building Regression Model using Keras: Part 2

In part 2, the network properties will again be increased to achieve a higher model accuracy. Later, the model will be saved, load and predict new concrete strength with new user-defined parameters.

Credit: IBM Cognitive Class

Building a Concrete Compressive Strength Model using Keras Framework

import pandas as pd
import numpy as np

url = 'https://s3-api.us-geo.objectstorage.softlayer.net/cf-courses-data/CognitiveClass/DL0101EN/labs/data/concrete_data.csv'

df = pd.read_csv(url)

df.head()

	Cement	Blast Furnace Slag	Fly Ash	Water	Superplasticizer	Coarse Aggregate	Fine Aggregate	Age	Strength
0	540.0	0.0	0.0	162.0	2.5	1040.0	676.0	28	79.99
1	540.0	0.0	0.0	162.0	2.5	1055.0	676.0	28	61.89
2	332.5	142.5	0.0	228.0	0.0	932.0	594.0	270	40.27
3	332.5	142.5	0.0	228.0	0.0	932.0	594.0	365	41.05
4	198.6	132.4	0.0	192.0	0.0	978.4	825.5	360	44.30

MODEL - Construct Model to Predict and Forecast.

Split Data to Predictors and Target

X = df.iloc[:,:-1]
X.head()

	Cement	Blast Furnace Slag	Fly Ash	Water	Superplasticizer	Coarse Aggregate	Fine Aggregate	Age
0	540.0	0.0	0.0	162.0	2.5	1040.0	676.0	28
1	540.0	0.0	0.0	162.0	2.5	1055.0	676.0	28
2	332.5	142.5	0.0	228.0	0.0	932.0	594.0	270
3	332.5	142.5	0.0	228.0	0.0	932.0	594.0	365
4	198.6	132.4	0.0	192.0	0.0	978.4	825.5	360

y = df.iloc[:,-1]
y.head()

0    79.99
1    61.89
2    40.27
3    41.05
4    44.30
Name: Strength, dtype: float64

Normalizing Data

X_norm = (X - X.mean()) / X.std()
X_norm.head()

	Cement	Blast Furnace Slag	Fly Ash	Water	Superplasticizer	Coarse Aggregate	Fine Aggregate	Age
0	2.476712	-0.856472	-0.846733	-0.916319	-0.620147	0.862735	-1.217079	-0.279597
1	2.476712	-0.856472	-0.846733	-0.916319	-0.620147	1.055651	-1.217079	-0.279597
2	0.491187	0.795140	-0.846733	2.174405	-1.038638	-0.526262	-2.239829	3.551340
3	0.491187	0.795140	-0.846733	2.174405	-1.038638	-0.526262	-2.239829	5.055221
4	-0.790075	0.678079	-0.846733	0.488555	-1.038638	0.070492	0.647569	4.976069

Save number of feature columns, n_cols to use later in model development.

n_cols = X.shape[1]
n_cols

8

Importing Libraries

from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score

import keras

from keras.models import Sequential
from keras.layers import Dense

Using TensorFlow backend.

Building the Model

Network Properties:

• Hidden Layers: 7
• Nodes: 25
• Activation Function: ReLU
• Optimizer: Adam
• Loss Function: Mean Squared Error
• Epochs: 250

mse = []
r2 = []

for i in range(50):
    
    #Split Data to Train and Test Set
    X_train, X_test, y_train, y_test = train_test_split(X_norm, y, test_size = 0.3)

    #Create model
    model = Sequential()
    model.add(Dense(25, activation='relu', input_shape=(n_cols,)))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(25, activation='relu'))
    model.add(Dense(1))

    #Compile model
    model.compile(optimizer='adam', loss='mean_squared_error')

    #fit the model
    model.fit(X_train, y_train, epochs=250, verbose=0)

    #predict output on test set
    y_pred = model.predict(X_test)
    
    mse.append(mean_squared_error(y_test, y_pred))
    r2.append(r2_score(y_test, y_pred))

print('mse_Mean: {:.2f}'.format(np.mean(mse)))
print('mse_StdDev: {:.2f}'.format(np.std(mse)))

mse_Mean: 31.73
mse_StdDev: 5.62

print('R^2_Mean: {:.2f}'.format(np.mean(r2)))
print('R^2_StdDev: {:.2f}'.format(np.std(r2)))

R^2_Mean: 0.89
R^2_StdDev: 0.02

from IPython.display import HTML, display
import tabulate

tabletest = [['MSE: Mean','MSE: StdDev','R^2: Mean','R^2: StdDev'],
         [round(np.mean(mse),2), round(np.std(mse),2), round(np.mean(r2),2), round(np.std(r2),2)]]

display(HTML(tabulate.tabulate(tabletest, tablefmt='html')))

MSE: Mean	MSE: StdDev	R^2: Mean	R^2: StdDev
31.73	5.62	0.89	0.02

Comparing the results from Part 1:

• mean squared error has gone down from 112.99 to 31.73
• R^2 has gone up from 0.59 to 0.89

which means the overall accuracy has gone up compared from the previous run.

SAVING MODEL

model.save('keras_reg.h5')

Predict New Concrete Strength Value with Pretrained Model

Import Module

import numpy as np
from keras.models import load_model

Load Pretrained Model

keras_reg = load_model('keras_reg.h5')

Insert New Parameters

#Cement
x1 = 1540.0

#Blast Furnace Slag
x2 = 250.0

#Fly Ash
x3 = 0.0

#Water
x4 = 162.0

#Superplasticizer
x5 = 2.5

#Coarse Aggregate
x6 = 1055.0

#Fine Aggregate
x7 = 676.0

#Age
x8 = 450

X = np.array([[x1, x2, x3, x4, x5, x6, x7, x8]])
X_norm = (X - X.mean()) / X.std()
strength_pred = keras_reg.predict(X_norm)

New Prediction

print('The predicted concrete strength is {:.2f}'.format(strength_pred[0][0]))

The predicted concrete strength is 56.26