An AI to recognize handwritten digits
Using tensorflow's, neural networks, this model identifies handwritten digits with an accuracy of 98.25%.
Inside handwriting.py is the structure for the neural network used that currently performs as so:
Epoch 1/10
1875/1875 [==============================] - 21s 11ms/step - loss: 0.2905 - accuracy: 0.9134
Epoch 2/10
1875/1875 [==============================] - 23s 12ms/step - loss: 0.0930 - accuracy: 0.9741
Epoch 3/10
1875/1875 [==============================] - 21s 11ms/step - loss: 0.0658 - accuracy: 0.9827
Epoch 4/10
1875/1875 [==============================] - 20s 11ms/step - loss: 0.0511 - accuracy: 0.9856
Epoch 5/10
1875/1875 [==============================] - 24s 13ms/step - loss: 0.0386 - accuracy: 0.9890
Epoch 6/10
1875/1875 [==============================] - 20s 11ms/step - loss: 0.0307 - accuracy: 0.9913
Epoch 7/10
1875/1875 [==============================] - 20s 11ms/step - loss: 0.0248 - accuracy: 0.9925
Epoch 8/10
1875/1875 [==============================] - 20s 11ms/step - loss: 0.0211 - accuracy: 0.9939
Epoch 9/10
1875/1875 [==============================] - 30s 16ms/step - loss: 0.0168 - accuracy: 0.9952
Epoch 10/10
1875/1875 [==============================] - 20s 11ms/step - loss: 0.0152 - accuracy: 0.9957
313/313 - 1s - loss: 0.0895 - accuracy: 0.9825
hr_model is a trained neural network that you can use with recognition.py to draw digits and classify them(unless you want to modify the handwriting.py in which case you'd have to resave the model).
MNIST is a dataset in TensorFlow that contains many sample handwritten digits that we can use to train our neural network(read more)