hey Marco!! This is the first time I've watched one of your videos, and after 5 minutes of starting the video, I quickly went through your entire channel, looking at your content. It's AMAZING! Thank you for all your efforts to share your knowledge with the community. A hug from Chile!!
Excellent presentation. Very clear explanation. Would be great to have more info on the impact of the context and wich one of the methods is expected to work best in wich context.
Hi Marco! I'm working on a project and this has a lot of components I need. I noticed the specification of the data said that it was being recorded every 5 minutes, could you create a tutorial on how to retrieve a stream of live data and pass it to the algorithm in a somewhat real-time fashion? I hope this is similar to what I understood from your data collection in the video
Here, accuracy is really not a good idea, because there are so few anomalies. A simple baseline could achieve 99% accuracy, even though there is no "learning". That's why we use the confusion matrix here to see if we can actually identify anomalies.
Little criticism, I didn't find your explanation of robust z-scores very clear.. You use the acronym MAD to mean two different things. When talking about the Median Absolute Deviation, you still have Mean Absolute Deviation displayed on the slide title. Then you talk about the Z-score when the formula shows M_i. I found that section confusing and went elsewhere to look it up. Otherwise thanks for the video! I learnt some new things :)
Content is good, but the title is pretty missleading. I was expecting anomaly detection in "TIME SERIES" but actually you are not taking into account the time at all :) Seasonality? Time? hour? day of the week? month?
hey Marco!! This is the first time I've watched one of your videos, and after 5 minutes of starting the video, I quickly went through your entire channel, looking at your content. It's AMAZING! Thank you for all your efforts to share your knowledge with the community. A hug from Chile!!
Thanks Pablo for the kind words! Really appreciate it!
You rock bro🎩 off to you.
Thanks for this
We Hope to make Some One For MultiVariate Time Series Anomaly Detection
Excellent presentation. Very clear explanation. Would be great to have more info on the impact of the context and wich one of the methods is expected to work best in wich context.
Anomaly detection is unsupervised, how did you get to if a point is anomaly or not, even before training the model ?
The dataset is labeled. That way, we can measure the performance of each anomaly detection methods.
We got a few positive labels in cross validation
Hello Marco, thank you so much for such a great video. Can you please make a video on anomaly detection for time series data using pycaret.
Hi! Do you recomend any video for pattern-wise anomaly detection?
I don't know any, but you can look at the library TOAD for anonaly detection in time series. They do pattern-wise detection if I remember well
Hey !, Is it possible to identify and flag anomalies within a continuous numerical attribute?
If by continuous, you mean at a very high frequency, then yes, I don't see why not!
Thanks !, If possible, can you make a video on that, it would be really helpful !@@datasciencewithmarco
Hello!! quick question, why is the threshold 3.5 any reason please?
Very interesting content, thank you!
Hi Marco!! Thank you so much for making great videos on "Anomaly detection". Great Great work! Please keep sharing! 🙏🙏🙏🙏
Hi Marco! I'm working on a project and this has a lot of components I need. I noticed the specification of the data said that it was being recorded every 5 minutes, could you create a tutorial on how to retrieve a stream of live data and pass it to the algorithm in a somewhat real-time fashion? I hope this is similar to what I understood from your data collection in the video
Hi I wanted to work on the same thing, did you get anything?
how about random cut forest ?
What is the accuracy?
Here, accuracy is really not a good idea, because there are so few anomalies. A simple baseline could achieve 99% accuracy, even though there is no "learning". That's why we use the confusion matrix here to see if we can actually identify anomalies.
nice and clear
Great video
🎉 thank you a lot
Thanks for this 🤘
Little criticism, I didn't find your explanation of robust z-scores very clear..
You use the acronym MAD to mean two different things. When talking about the Median Absolute Deviation, you still have Mean Absolute Deviation displayed on the slide title. Then you talk about the Z-score when the formula shows M_i. I found that section confusing and went elsewhere to look it up.
Otherwise thanks for the video! I learnt some new things :)
awesome!
Content is good, but the title is pretty missleading. I was expecting anomaly detection in "TIME SERIES" but actually you are not taking into account the time at all :) Seasonality? Time? hour? day of the week? month?