DIgit-Recognizer

This is the code for a Kaggle compitition called DIgit-Recognizer

I use a easy CNN to train this model

This is the structure of the model

---

Layer (type) Output Shape Param #

conv2d_28 (Conv2D) (None, 26, 26, 32) 320

---

activation_46 (Activation) (None, 26, 26, 32) 0

---

batch_normalization_28 (Batc (None, 26, 26, 32) 104

---

max_pooling2d_28 (MaxPooling (None, 13, 13, 32) 0

---

zero_padding2d_19 (ZeroPaddi (None, 15, 15, 32) 0

---

conv2d_29 (Conv2D) (None, 13, 13, 48) 13872

---

activation_47 (Activation) (None, 13, 13, 48) 0

---

batch_normalization_29 (Batc (None, 13, 13, 48) 52

---

max_pooling2d_29 (MaxPooling (None, 6, 6, 48) 0

---

zero_padding2d_20 (ZeroPaddi (None, 8, 8, 48) 0

---

conv2d_30 (Conv2D) (None, 7, 7, 64) 12352

---

activation_48 (Activation) (None, 7, 7, 64) 0

---

batch_normalization_30 (Batc (None, 7, 7, 64) 28

---

max_pooling2d_30 (MaxPooling (None, 3, 3, 64) 0

---

dropout_10 (Dropout) (None, 3, 3, 64) 0

---

flatten_10 (Flatten) (None, 576) 0

---

dense_19 (Dense) (None, 3168) 1827936

---

activation_49 (Activation) (None, 3168) 0

---

dense_20 (Dense) (None, 10) 31690

---

activation_50 (Activation) (None, 10) 0

after 100 Epoch of training we get the best train acc of 100.00%, the best vall acc of 99.97%

And we get the acc of in Kaggle test set