Hi, I just read your comment and I wanted to know how's your data science career going? I just completed ML and going to create an ML project for resume. Can you please give me any kind of suggestion if you are reading this comment.

Hey Have you become data scientist ? If yes please suggest me something

can u please explain ur outliers point

@@crackthecode1372 outliers are just noise

@@lars1597 Sometimes outliers are important noise. Outliers can tell a lot about data. They can't simply be dropped.

Minmax scaler is the most widely used in forecasting research papers. Z-score is not very good in time series forecasting

Thanks

hello , i want to learn statistics for data science i don't have prior knowledge, will this cover the basics as i want to start from scratch

Me too moving for the next Playlist that is Feature Engineering... Good Luck

@@matinpathan5186 Good luck to you too

can u please explain ur point?

In Standardization, values can theoretically go outside -3 and + 3 also. They will lie in .15% of values in each tail. Basically, the outliers.

Hey. Data collection or Data Acquisition is one of the very first steps in designing a data science pipeline. Often Data Scientists working in tech companies get easy access to this data but if you want to collect data for yourself (on a small scale) then stuff like web scraping comes into play. Web scraping is nothing but collecting data from a website. Web Scraping is a popular means of collecting data. Often companies survey people to collect data. It can be an online survey or an offline one. The means of collecting data are endless.

@@RitwikDandriyal thank you for the reply.. 😊

Bro video about difference between fit, transform and fit_transform !!

yes i was also searching for the GitHub link .Why is it missing??

Yes u r right OLS takes care of it

Had the same question. Turned out he made a mistake, StandardScaler calculates mu and sd for the dataset you fit into it. Next time have similar question - go right to the documentation. StandardScaler - scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html.

Hi Sejal, the topic you are asking will not fall under feature scaling. As you are using the one hot encoding it will result in large no.of columns, instead of using one hot encoding you can use some other encoding technique such as feature encoding,target encoding,binary encoding,k-fold target encoding. In these encoding techniques every one has its own pros and cons, you have to compromise in some of the things and you have to do the modelling

@@bonalareddy5339 okay so you mean instead of feature encoding i apply other encoding techniques and then scale the dataset?

@@sejalchandra2114, every feature not needed to be scaled. The feature that are to be scaled are the numerical one's which doesn't fall in the same range and same unit, as categorical features are converted into some meaningful numerical feature with the help encoding techniques and those numerical formats(in general) doesn't affect the model that much. For suppose there are 2 numerical features f1, f2. f1 has a range btw 10 to 1000 and f2 has a range btw 50 to 300 in this case the features are needed to be scaled as those both features are different and the units f1, f2 are talking might be different. f1 might be talking weight in kgs, f2 might be talking about height in cms, if you are using anyone of the modelling algorithm such as linear regression, logistic regression, knn, k means, cnn, ann(as mentioned in the video)you are supposed to do the feature scaling. If you are using some other techniques such as decision trees, random forest or anyother bagging/boosting techniques there is no point in scaling the features as it doesn't affect the performance of the algorithm(in case of tuning or faster output, also mentioned in the video). If you feel that for dealing with the categorical variables one hot encoding is producing a large no.of columns you can use some other encoding (that's what i meant in my previous reply). I hope this helps.

@@bonalareddy5339 okay thanks I got the point. Really appreciate your detailed response.