Azure Machine Learning Studio: Introduction to prediction
HTML-код
- Опубликовано: 21 сен 2017
- Dataset: www.ishelp.info/data/bikebuyer...
This playlist (or related videos) is used in two of my online books:
1. Data Analytics and Machine Learning
- Contents: app.myeducator.com/reader/web...
- Purchase: app.myeducator.com/s/1eKHmUCl...
2. Data-Mining Projects and Database Essentials
- Contents: app.myeducator.com/reader/web...
- Purchase: app.myeducator.com/s/2NgsEMdh...
To request an instructor copy, email me at mark.keith@gmail.com
Great videos! Thank-you for posting!
Thanks dude, this is exactly what I was looking for, being a noob.
Good vids. Thanks! FYI, I wanted to address the issue with column selection that is a recurring PITA in these for you. It's because of the event handling (or mis-handling). Workaround: Just click outside the textbox when there's no "Press enter to enter column name" in that textbox (when the dropdown is "column names"). And Microsoft, FYI, that's how you'd fix it. Check for that value on the change event of the dropdown when going to "column names". Geez! But it is tricky to get these changes up the flagpole, so given it's been 2+ years and I am having this issue as a newbie today, years since you had issues with it, maybe this can be helpful. Once you get the pattern down, it's less frustrating and works fine and reliably. HTH and again, thanks. Nice and useful vids!
Insta-subbed, great content, well-explained, mistakes during the process make it even better because of the solutions and explanations you provide for the process flow.
Thank you for this content, I will definitely check the other videos, keep up the good work!
Thanks!
Correct if I'm wrong, but you're predicting a Yes/No 0/1, which is a binary classification. Logistic regression should do the job. Don't quite understand why you use Linear Regression.
Wouldn't it be better to use gradient descent instead of linear regression since we have multiple variables?
Hello, thank you for a clearest and straightforward explanations.
Is there a way to change the threshold of the prediction ? I guess Azure uses a 0.5 threshol.
Thank you,
Tamir
I am looking for guidance on how to approach a specific experiment. Where do I find individuals or firms that help with this? I dont want them to do it for me I just want them to guide me.
I probably misunderstand the meaning of the R square score but...isn't 0.5 stands for random prediction? Meaning we got here 0.101 or 0.104 ...
Does it mean that the model is no good?
Thank you,
Tamir
Good question! 0.10 R squared means that the features (independent variables) can explain 10% of the variance in the labels (dependent variables). In this dataset, you're right that we have a 50/50 chance of predicting correctly whether or not they purchased a Bike. However, a 10% R squared doesn't equate to being right 10% of the time. Rather, a 0% R squared means that you'd correctly guess the outcome 50% of the time (since there are two possible outcomes).
Therefore, think of the R squared in this video more like being 10% better than a random guess. That's not a perfect definition because it doesn't necessarily mean we'll be right 60% of the time. But conceptually, that's a better way of explaining it. Does that make sense?
Thank you for the answer.
I think I mixed up with the AUC value where 0.5 means that the algorithm produces random predictions (I hope I got this one right).
Your explanation of the R-Square is clear to me.
Thank you,
Tamir
Tamir Basin Yep, that sounds right about area under the curve 👍🏼
Thanks for the video! The link to the dataset doesn't work anymore. Can you update it please?
Thanks; sorry about the dataset, I’ve been recreating my website and I don’t have it live quite yet. You’re welcome to email me if you’d like and I can send it to you directly
@@MarkKeith raw.githubusercontent.com/Apress/predictive-analytics-w-msft-azure-ML/master/Ch05-BikeBuyer.csv
Hello, everytime I evaluate my model, I get
Mean Absolute Error 0.000001
Root Mean Squared Error 0.000002
Relative Absolute Error 0.000003
Relative Squared Error 0
Coefficient of Determination 1
Can you please tell me what I have done wrong?
It sounds to me like one of your features is perfectly related to your label. In other words, that would be like predicting the numeric version of “Purchased Bike” with the text-based version of “Purchased Bike”. The result is a 100% R squared value (i.e. coefficient of determination = 1). Could that be the problem? If so, just make sure that you eliminate the text-based version of Purchased Bike using the “Select Columns” pill
Thank you for the repose. Would work on that
is it able to predict stocks ?
Yes, but you need to use a forecast algorithm. I haven’t made a video for that yet. Sorry!