Briefly
I created a new project, Interactive Machine Learning Experiments on GitHub. Each experiment consists of a Jupyter / Colab laptop showing how the model was trained, and a Demo page showing the model in action right in your browser.
Despite the fact that the machine models in the repository can be a little "dumb" (remember, these are just experiments, not a licked code, ready for "uploading to production" and further managing new Tesla), they will try their best to:
- Recognize numbers and other sketches that you draw in the browser
- Identify and recognize objects in video from your camera
- Categorize images uploaded by you
- Write a poem in the style of Shakespeare with you
- And even play rock-paper-scissors with you
- and so forth
I trained models in Python using TensorFlow 2 with Keras support . For the demo application, I used the React and JavaScript version of Tensorflow .
![Interactive Machine Learning Experiments](https://habrastorage.org/getpro/habr/post_images/3e4/b83/2a6/3e4b832a607b38fe6ebf64a74304925e.png)
Performance models
, .️ , " ", . , . (, 60% , , 97%), (overfitting vs underfitting).
- :
![Dumb machine model](https://habrastorage.org/getpro/habr/post_images/079/73c/c81/07973cc81709f3ceb90984bf76ea68b7.png)
, - "":
![Smarter machine model](https://habrastorage.org/getpro/habr/post_images/f0c/246/e87/f0c246e87d8461bedaac585f7531b9b0.png)
- full-stack (-). , , , - .
- Python , . Playground and Cheatsheet for Learning Python. , Python, "" ,
dict_via_comprehension = {x: x**2 for x in (2, 4, 6)}
. - Python- . Andrew Ng Coursera Homemade Machine Learning. " -" , , , k-, ( ?) .
- " " NanoNeuron. 7 JavaScript , , - "".
- Deep Learning - Andrew Ng Coursera (multilayer perceptrons), (convolutional and recurrent neural networks). , , . TensorFlow 2 Keras. ( ), - , -, . Interactive Machine Learning Experiments, .
- TensorFlow.js , ( Jupyter ) .
- HDF5 TensorFlow.js Layers TensorFlow.js converter. ( ) , HTTP , , -, , , , "". HTTP API .
- React create-react-app Flow .
- Material UI. , , " " (, Bootstrap).
- , Jupyter :
(Multilayer Perceptron, MLP)
, .
![Handwritten Digits Recognition](https://habrastorage.org/getpro/habr/post_images/dde/abb/820/ddeabb820e5eaf3a6737ea3798805b22.gif)
, .
![Handwritten Sketch Recognition](https://habrastorage.org/getpro/habr/post_images/a1a/f98/f24/a1af98f24f9687516f436c3997306916.gif)
(Convolutional Neural Network, CNN)
(CNN)
, . MLP, CNN.
![Handwritten Digits Recognition (CNN)](https://habrastorage.org/getpro/habr/post_images/dde/abb/820/ddeabb820e5eaf3a6737ea3798805b22.gif)
(CNN)
, . MLP, CNN.
![Handwritten Sketch Recognition (CNN)](https://habrastorage.org/getpro/habr/post_images/a1a/f98/f24/a1af98f24f9687516f436c3997306916.gif)
-- (CNN)
-- . CNN, .
![Rock Paper Scissors (CNN)](https://habrastorage.org/getpro/habr/post_images/ba1/ce2/c40/ba1ce2c409ef1f859f2379aa913b884a.gif)
Rock Paper Scissors (MobilenetV2)
-- . , MobilenetV2.
![Rock Paper Scissors (MobilenetV2)](https://habrastorage.org/getpro/habr/post_images/62b/d44/fd2/62bd44fd2baf2044c85dcf300ca340e2.gif)
(MobileNetV2)
( ), . MobilenetV2.
![Objects Detection (MobileNetV2)](https://habrastorage.org/getpro/habr/post_images/85e/cee/fa3/85eceefa39aff6781db023fe14c41d3e.gif)
(MobileNetV2)
, , "" . MobilenetV2.
![Image Classification (MobileNetV2)](https://habrastorage.org/getpro/habr/post_images/687/de9/629/687de96293f15861031ad7fc1b6da1ed.gif)
(Recurrent Neural Networks, RNN)
(, 17+38
) ( 55
). , , ( ). "" 1
→ 17
→ 17+
→ 17+3
→ 17+38
55
. , Hola
Hello
, .
![Numbers summation](https://habrastorage.org/getpro/habr/post_images/e49/3ba/110/e493ba110b3512b3c8367d0c1287905f.gif)
, . "" .
![Shakespeare text generation](https://habrastorage.org/getpro/habr/post_images/93e/bae/a4c/93ebaea4c757d34a8d9ba02e705fcaa9.gif)
Wikipedia
Wiki , .
![Wikipedia Text Generation](https://habrastorage.org/getpro/habr/post_images/9ef/c31/06b/9efc3106b06044a6e4f6ff91e2d5cbbc.gif)
, — , (, ). — Deep Learning. :
, . , / , 60%
.
, , - , --.
!