https://github.com/jefferyustc/mnistonkeras

Implemention of lightweight CNN Model on Mnist Dataset,try to include as more as possible features that I learn from keras, by the way,compare to tensorflow.hope to give a good learning example.
https://github.com/jefferyustc/mnistonkeras

deep-learning keras-tensorflow mnist-classification

Last synced: 2 months ago
JSON representation

Implemention of lightweight CNN Model on Mnist Dataset,try to include as more as possible features that I learn from keras, by the way,compare to tensorflow.hope to give a good learning example.

• Host: GitHub
• URL: https://github.com/jefferyustc/mnistonkeras
• Owner: jefferyUstc
• License: gpl-3.0
• Created: 2018-12-06T13:41:34.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2022-11-21T21:36:47.000Z (about 2 years ago)
• Last Synced: 2023-10-20T05:40:44.275Z (over 1 year ago)
• Topics: deep-learning, keras-tensorflow, mnist-classification
• Language: Python
• Size: 12.5 MB
• Stars: 5
• Watchers: 1
• Forks: 1
• Open Issues: 4
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# MnistOnKeras

Implemention of lightweight CNN Model on Mnist Dataset,try to include as more as possible features that I learn from keras, by the way,compare to tensorflow.hope to give a good learning example.

# Keras_workflow

  I used to using tensorflow, recently, decide to pay more attention to keras.so it's a guide example which shows how to train a model by keras.I found training a model by keras can be done in six steps:

1.  prepare a Dataset

2. build a model

3. compile the model(`compile()`)

4. train the model(`fit()`)

5. validate the model(`valuate()`)

6. optionly, predict on new data(`predict()`)

# Train_my_model

- train

![](./images/acc.png)

 

![](images/loss.png)

- validation

![](images/val_acc.png)

 

![](images/val_loss.png)

 

- net-structure

![](images/graph_run=.png)

# Contact

  Live and Learn, happy to exchange study, so any question, you can contact me: [email protected]