Was good. Thank you!

Un très bon contenu

Awesome Experience.

a bit complicated

Amazing to learn

nice intuition

very valuable

Excellent One

Good Course

good class

Marvelous

thank you

따라하기가 어렵다

Great

goood

great

good

GOOD

good

OK!!

good

v

I've spent the past three years studying ML and AI starting from the ground up with Calculus, Linear Algebra, basic data science techniques and eventually Deep Learning. I am primarily interested in this specialization because I would like to begin using GCP professionally. This course provides a very quick surface level overview of the "history" of ML and the techniques that have been aggregated to make up the current cutting edge of AI in practice. Already having a grasp on many of the concepts, I was able to zip through this course in a few hours and found it basic. If you're looking for something a bit more challenging, I would recommend the DeepLearning.ai specialization also available on Coursera. This course works well as a refresher and a high level overview. If you are completely new to the field, be warned that there is quite a terminology to be unpacked that is covered more thoroughly in other courses on Coursera. The University of Washington machine learning specialization (though sadly cut short) would be a much better starting place, if you are completely new to the topic.

Contingency tables and ROC graphs were poorly characterized and presenter resorted to obfuscation to mask his unfamiliarity with this basic statistical concept. Furthermore, when the proposed task is to "Identify pictures containing house cats", correctly identifying a picture that does not contain a house cat (True Negative) does NOT count as a successful prediction. You are confusing sensitivity with specificity in your so-called confusion matrix.

With respect to labs, you should warn students to leave their notebooks open so we do not have to reload everything. Also in the cab fare exercise the presenter did not elaborate on the fact that the RMSE's were higher than the predicted fare and mistakenly excluded time of day when in fact fares increase during rush hour.

Using hash function doesn't seem a good way to split the dataset:

-You could discard a relevant feature

-You will group data on a similar characteristic, which might not represent the population well

-You don't have control over the size of your split since the feature will not likely be uniformly distributed

Can't we add an index feature/column and do a modulo on the index?