Handwritten Digits Classification: Logistic Regression from Scratch in Python 

Hi Mohamad, thank you for your question. The Digits Dataset I am using from Sklearn is has pixel integer values of 0-16, I am dividing by 16 to normalize the data and make the numbers more manageable numerically,

@@EndlessEngineering Thank you so much for your help and attention

Hi Samuel, Thanks for watching the video. The link for logistic regression and gradient descent concepts videos are provided in the "Suggested Video" section, that is the letter "i" at the top right of the video. There is also a link to the video for logistic regression in the jupyter notebook that I haves linked in the description.

@@EndlessEngineering thank you

Hi Anonimo, thanks for you question. The syntax I am using when splitting the data is "=\" and not "=/". The "\" can be used in python to split a line into two, in this case I am splitting the line after the assignment operator. It is just to make all the code fit on screen, but if I don't use the "\" syntax python would throw an error.

@@EndlessEngineering Wow, thanks for the answer in such short time. This was really my first time seeing this application. cheers

The "\" is used to indicate line continuation to the python interpreter. Basically if you each a long line and you want to break it up you can use "\" to do so

Hi Karine, thanks for your question. Mathematically you are correct we would need to transpose the params vector to get the dot product. However, the np.dot() functionality allows for doing it without the transpose, as in the documentation it shows that if you have 1-D arrays (vectors) the result is always the inner (dot) product. See np.dot documentation here: numpy.org/doc/stable/reference/generated/numpy.dot.html

Sir can u plz tell me as soon as possible

@@kainnatbatool7021 Hi Kainnat, can you please share what error you are getting? Are you using the jupyter notebook here:github.com/endlesseng/ml-vid-code/blob/master/notebooks/logistic_regression_digits_dataset.ipynb

You could think about it like that, although mathematically it is not the similar to a neural network exactly since there is no network with inner connections here. If you were to make a network of logits (on of the models I train in this video) then you can get something similar to a neural network.

Each classifier has as many params as there are pixels in each image.