MNIST_on_Tensorflow
Data is taken from kaggle having 32000 samples.
- Training Size: 30000
- Testing Size: 2000
- Accuracy: 99.3%
| Network Configuration | Filter Size | #Filters | Max Pooling/ReLU |
|---|---|---|---|
| Convolutional Layer 1 | 5x5 | 16 | Yes |
| Convolutional Layer 2 | 5x5 | 32 | No |
| Convolutional Layer 3 | 2x2 | 64 | Yes |
| Convolutional Layer 4 | 2x2 | 128 | No |
| Fully Connected Layer 1 | N/A | 512 | Yes |
| Fully Connected Layer 2 | N/A | 256 | Yes |
| Fully Connected Layer 3 | N/A | 10 | No (Logistic Layer) |
- Hardware: NVIDIA GeForce 670 Ti OEM
- Running Time: 6-7 Minutes
dataset.py: for dataset formatting and extract batches of training samples at each iterations.
Testing set is extracted from train.csv only.
Ref: https://github.com/Hvass-Labs/TensorFlow-Tutorials/blob/master/02_Convolutional_Neural_Network.ipynb