Introduction

For this assignment, I have implemented logistic regression for image files of digits written by hand from MNIST data-set to classify the digits written from 0 to 9. The model is used to find the probability of each digit written, if it is reasonable like more than 75% then it is possible to accept it as the predicted digit.

Objectives

To find the classification accuracy of logistic regression technique for classifying hand-written numbers to a number.

Approaches/Methods

• First, I downloaded the required data-set • Next, I built the computation graph • Then, model function and cost function are evaluated • Finally, I trained and evaluated the model

Datasets

The data set is composed of three parts: • Points of training data • Points of testing data, and • Points of validation data

Workflow

Parameters

Learning Rate = 0.075 Epochs = 25 Batch Size = 100 Display Step = 2 Evaluation Epoch: 0001 cost= 0.596986147 Epoch: 0002 cost= 0.384767043 Epoch: 0003 cost= 0.350911226 Epoch: 0004 cost= 0.333020710 Epoch: 0005 cost= 0.321855854 Epoch: 0006 cost= 0.313655997 Epoch: 0007 cost= 0.307405996 Epoch: 0008 cost= 0.302446134 Epoch: 0009 cost= 0.298442693 Epoch: 0010 cost= 0.295032452 Epoch: 0011 cost= 0.291788620 Epoch: 0012 cost= 0.289517276 Epoch: 0013 cost= 0.287142176 Epoch: 0014 cost= 0.285069724 Epoch: 0015 cost= 0.283137434 Epoch: 0016 cost= 0.281469407 Epoch: 0017 cost= 0.279919414 Epoch: 0018 cost= 0.278679141 Epoch: 0019 cost= 0.277288633 Epoch: 0020 cost= 0.276062239 Epoch: 0021 cost= 0.275074296 Epoch: 0022 cost= 0.273868422 Epoch: 0023 cost= 0.272837362 Epoch: 0024 cost= 0.271818635 Epoch: 0025 cost= 0.271095491

Conclusion

As the result, we trained a regression-model to look at the pictures of numbers written by hand and predict what they really are.

Output