- 安装更新
TensorFlow pip包，并验证
pip install --upgrade tensorflow
python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000]TensorFlow快速入门与实战
| 章节 | 深度学习框架 | 笔记博客 | 案例代码 |
|---|---|---|---|
| 01 | tensorflow入门 | 01-tensorFlow入门.ipynb 02-手写体数字识别.ipynb |
|
| 02 | 基本概念 | 01-张量初体验.ipynb 02-变量.ipynb 03-操作.ipynb 04-会话.ipynb |
|
| 03 | 房价预测 | 01-数据分析.ipynb 02-数据规范化.ipynb 03-创建数据回归模型.ipynb 04-TensorBoard名字作用域.ipynb 05-可视化损失函数.ipynb |
|
| 04 | 手写体数字识别 | 01-加载MNIST数据集.ipynb 02-MNIST-softmax.ipynb 03-MNIST-CNN.ipynb |
|
| 05 | 验证码识别 | ||
| 06 | 人脸识别 |
this is the homework according to YouTube online courses of Prof. Lee from National Taiwan University
veeeery understandable and clear https://www.youtube.com/playlist?list=PLJV_el3uVTsPy9oCRY30oBPNLCo89yu49
i took some notes which i think it's interesting and new for me, and it's in English, not very detailed, but include important conclusions i got from the course, hope you could find something useful.
HW2 is about using education, nationality and other features to predict the incomes. I both tensorflow, and keras... the accuracy on validation data is about ~86% Cheers!
HW3 is doing the sentiment classification, the descripiton of HW, and data link can be found fromhttps://ntumlta.github.io/ML-Assignment3/index.html I also upload the notes i took in lectures :
- why we using CNN for image processing?
- tips for DNN (how to improve the model by analysis the performance on training and validation data)
- Why Deep? (why not wide) (because of memrory problem, the result is not good [bad, i would say (o(╥﹏╥)o)]), if you have bigger memory, you could deeper the network, which I think is quite simple using keras.
If you have the same problem like me, you could utilize the tips in "Tipps for DNN" to improve the result in HW2.
Cheers!
