arogozhnikov / arogozhnikov.github.io Goto Github PK

This is a comment about your results from Reconstructing pictures with ML
NNs are not too bad if you make them bigger and use ReLU instead of Tanh (using Karpathy's demo):

An interesting question is how well the other ML methods scale compared to NNs.

may you share function load_problems
so this line will not give error
import load_problems
then it it will be possible to reproduce results
thanks
by the way , what you think why logistic regression is so good
actually there is no need in FM??

Would love to see it if you still have it.

Thank you for a great post on comparison of different Factorization Machine implementations.

I have a very basic query; Is factorization machine designed to work only with binary fields? Do we need to one hot encode all features? How are real-valued featured handled?

Thank you!

На странице http://arogozhnikov.github.io/2016/06/24/gradient_boosting_explained.html в Хроме пишет, что "Webgl is not supported by your browser - visit http://get.webgl.org for more info". Но зайдя на http://get.webgl.org я вижу вращающийся куб и написано: "Your browser supports WebGL", т.е. он всё таки работает.

В консоле браузера выводится следующая ошибка:

plotly-1.13.js:77919 Uncaught TypeError: Cannot read property 'canvas' of undefined
initializeGLPlot@ plotly-1.13.js:77919
Scene @ plotly-1.13.js:78005
plotGl3d @ plotly-1.13.js:76899
drawData @ plotly-1.13.js:63735
lib.syncOrAsync @ plotly-1.13.js:61329
Plotly.plot @ plotly-1.13.js:63772
Plotly.newPlot @ plotly-1.13.js:64297
DecisionTreeVisualization @ gradient_boosting_explained.js:143
(anonymous function) @ gradient_boosting_explained.js:175

Видимо или ошибка в plotly или в том, как его используют.

In notebooks/2015-09-30-NumpyTipsAndTricks2.ipynb,

def running_average_simple(seq, window=100):
    result = np.zeros(len(seq) - window)  # The bug is in this line
    for i in range(len(result)):
        result[i] = np.mean(seq[i:i + window])
    return result

Should be

def running_average_simple(seq, window=100):
    result = np.zeros(len(seq) - window + 1)  # Fixed off by one bug
    for i in range(len(result)):
        result[i] = np.mean(seq[i:i + window])
    return result

hi, happy to read your post "Learning to rank (software, datasets)"
In this post , you say the text of query and document are available of MSLR-WEB10k and MSLR-WEB30k,

However, I couldn't find the text using this download link https://www.microsoft.com/en-us/research/project/mslr/?from=http%3A%2F%2Fresearch.microsoft.com%2Fen-us%2Fprojects%2Fmslr%2Fdownload.aspx.

Could you help me to tell the accurate download link.

Thank you!