Convolutional Neural Networks

I have meticulously configured a Convolutional Neural Network with the aid of TensorFlow.

This sophisticated design encompasses convolutional layers, max-pooling, and judiciously implemented dropout techniques. This is further complemented by fully-connected layers. The model underwent training with batches of 128, over a span of 30 cycles.

So without wasting time let’s get start

---

import os
os.environ['KMP_DUPLICATE_LIB_OK']='True'

from sklearn import datasets, metrics
from sklearn.model_selection import train_test_split
import numpy as np
import tensorflow as tf

print('Tensorflow version:',tf.__version__)

tf.random.set_seed(2)
np.random.seed(0) # using this to help make results more reproducible
tf.config.list_physical_devices('GPU')

digits = datasets.load_digits()

# NORMALIZATION, SO IMPORTANT
X = digits.data.astype(np.float32)/16.0 - 0.5
y = digits.target.astype(np.int32)

X = X.reshape((X.shape[0],8,8,1)) # reshape as images

# Split it into train / test subsets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

# Split X_train again to create validation data
#X_train, X_val, y_train, y_val = train_test_split(X_train,y_train,test_size=0.2)

X_train.shape

In our latest experiment, we started off by pulling the digits dataset straight from sklearn. Once we had that, we took the time to carefully normalize our data. You know how crucial that is, right? Then, we divvied it up: one chunk for training and another for testing. To fit the image processing mold, we had to reshape our data a bit. Oh, and by the way, we did check the TensorFlow version we’re working with and also took a peek at the size of our training batch.