Machine Learning - MNIST digit prediction
MNIST Neural Network Digit Recognizer: Built logistic classifier on MNIST digit image dateset. Neural network was built to classify the image.
Libraries: PyTorch , Keras
DataSet: 60,000 small square 28×28 pixel grayscale images of handwritten single digits between 0 and 9
Results:
Logistic: Misclassified samples: 947, Accuracy: 0.84
Neural Network: model with 2 hidden layers of size 100 and 25 (Baseline: 96.53% (0.37%))
Network Architecture - Layers Size [784 -> 512 ->256 -> 128 -> 64 -> 10] 0.9742857142857143 (97%)
Read More to know full details of project and code can be found on git link.
if unable to access github notebook , here is link to my colab (github has this bug going for quite long):
https://drive.google.com/file/d/1Q13AdZDRN3XJl4_T3ssiuiJbYkEtc6l5/view?usp=sharing
The MNIST handwritten digit classification problem is a standard dataset used in computer vision and deep learning. I picked this project as part of my academics to build the basis for learning and practicing how to develop, evaluate, and use convolutional deep learning neural networks for image classification from scratch.
The MNIST dataset is an acronym that stands for the Modified National Institute of Standards and Technology dataset.
Project Tasks :
1. EDA and Data Cleaning ( Standarize the dataset)
2. Model Building
3. Hyper-parameter Tuning
4. Predicting on Test Data
Once the basic model classes were ready, I experimented with different model sizes. Results of all the layer structures are as:
Results:
Logistic: Misclassified samples: 947, Accuracy: 0.84
Neural Network: model with 2 hidden layers of size 100 and 25 (Baseline: 96.53% (0.37%))
Network Architecture - Layers Size [784 -> 512 ->256 -> 128 -> 64 -> 10] 0.9742857142857143 (97%)