These notebooks apply various techniques taught in the Deep Learning Specialization at Coursera to the MNIST dataset. The goal is to explore hyperparameter tuning and various optimization techniques and their effects on the run time and accuracy of the ML model. The notebooks are implemented using NumPy to delve into the details and understand the ML models at a fundamental level.
As a reference, the first couple of notebooks implement the TensorFlow tutorials for beginners using Softmax logistic regression and for experts using Multilayer Convolutional Network.
These note books are all trained on a MacBook Pro (Retina, 13-inch, Late 2013) with a 2.6 GHz Intel Core i5 and 8 GB 1600 MHz DDR3 and no GPU acceleration.
| Notebook | Epochs | Train Time (Secs) | Train Acc | Val Acc | Test Acc | Comment |
|---|---|---|---|---|---|---|
| MNIST_TF_Softmax_Tutorial | 5 | 3.1157 | 0.9248 | 0.9232 | 0.9246 | Tensor Flow and Softmax Logistic Regression |
| MNIST_TF_CNN_Tutorial | 7 | 1781.1032 | 0.9965 | 0.9912 | 0.9896 | Tensor Flow and CNN |
| MNIST_NP_Softmax_Basic | 1000 | 659.55 | 0.9256 | 0.9274 | 0.9228 | Numpy and Basic Softmax Logistic Regression |