In machine learning, our main motive is to create a model that can relate the dependent variable(i.e target) with the independent variable(i.e. data). The most common model to do this is regression analysis. Regression fits the best possible curve on the training data set so that it can predict the target using the same curve.
So this recipe is a short example of How to perform basic regression using keras model?
import pandas as pd import numpy as np from keras.datasets import mnist from sklearn.model_selection import train_test_split from keras.models import Sequential from keras.layers import Dense from keras.layers import Dropout
We have imported pandas, numpy, mnist(which is the dataset), train_test_split, Sequential, Dense and Dropout. We will use these later in the recipe.
Here we have used the inbuilt mnist dataset and stored the train data in X_train and y_train. We have used X_test and y_test to store the test data.
(X_train, y_train), (X_test, y_test) = mnist.load_data()
We have created an object model for sequential model. We can use two args i.e layers and name.
model = Sequential()
Now, We are adding the layers by using 'add'. We can specify the type of layer, activation function to be used and many other things while adding the layer.
Here we are making regression model so we are making the linear stack of layers. We are using the activation function as 'relu' that is rectified linear unit, it has a advantage of being non linear also.
We can compile a model by using compile attribute. Let us first look at its parameters before using it.
model.compile(optimizer='Adam', loss='categorical_crossentropy', metrics=['accuracy'])
We can fit a model on the data we have and can use the model after that. Here we are using the data which we have splitted i.e the training data for fitting the model.
While fitting we can pass various parameters like batch_size, epochs, verbose, validation_data and so on.
After fitting a model we want to evaluate the model. Here we are using model.evaluate to evaluate the model and it will give us the loss and the accuracy. Here we have also printed the score.
score = model.evaluate(X_test, y_test, verbose=0)
print('Test loss:', score)
print('Test accuracy:', score)
Finally we are predicting the output for this we are using another part of the data that we get from test_train_split i.e. test data. We will use it and predict the output.
y_pred = model.predict(X_test)
As an output we get:
Epoch 1/2 469/469 [==============================] - 7s 14ms/step - loss: 0.3174 - accuracy: 0.9033 - val_loss: 0.1212 - val_accuracy: 0.9630 Epoch 2/2 469/469 [==============================] - 6s 14ms/step - loss: 0.1560 - accuracy: 0.9534 - val_loss: 0.0918 - val_accuracy: 0.9720 Test loss: 0.09184003621339798 Test accuracy: 0.972000002861023 [[8.92436292e-10 1.32853462e-09 6.39653945e-06 ... 9.99989152e-01 1.79315840e-09 2.44941958e-07] [9.11153306e-11 1.03196271e-05 9.99982357e-01 ... 1.89035987e-09 9.82423032e-09 8.40081246e-14] [1.10766098e-06 9.99514341e-01 1.26151179e-04 ... 1.44331687e-04 4.99823145e-05 6.05678633e-06] ... [2.03985762e-09 1.29704825e-08 2.95020914e-08 ... 1.23884201e-05 6.87194824e-06 1.75449488e-04] [5.91818647e-08 1.97798578e-08 7.46679774e-10 ... 5.06311437e-09 1.96506153e-04 1.14137793e-08] [1.13083731e-09 5.45665553e-12 2.54836174e-09 ... 3.70580059e-13 6.02386641e-10 3.15489106e-12]]