If you are a software developer who wants to build scalable AI-powered algorithms, you need to understand how to use the tools to build them. This Specialization will teach you best practices for using TensorFlow, a popular open-source framework for machine learning.
In this fourth course, you will learn how to build time series models in TensorFlow. You’ll first implement best practices to prepare time series data. You’ll also explore how RNNs and 1D ConvNets can be used for prediction. Finally, you’ll apply everything you’ve learned throughout the Specialization to build a sunspot prediction model using real-world data! The Machine Learning course and Deep Learning Specialization from Andrew Ng teach the most important and foundational principles of Machine Learning and Deep Learning. This new DeepLearning.AI TensorFlow Developer Specialization teaches you how to use TensorFlow to implement those principles so that you can start building and applying scalable models to real-world problems. To develop a deeper understanding of how neural networks work, we recommend that you take the Deep Learning Specialization.
WE
The course is fantastic. It was a bit short and with some hyperparameters tuning focus, it could have been great. Also, it seems that it is biased to show that LSTM is always superior to RNN networks.
VH
Coming from a background of knowing Deep Learning and theory of Time Series, this course was extremely helpful in understanding the practical aspects. I would recommend you take a course as well
SP
I really love this course. It's really easy to understand and systematic. please create more course about time series forecasting. Thank you very much. I've learned a lot of things from this course.
FF
This is a very suitable course for those of you who are new to machine learning, because after I took this course my interest in machine learning has increased. especially CNN computer vision.
BS
Brilliant end to a brilliant series.After taking this, one is well served to visit the DeepLearning.ai channel on YouTube and look at some of the more in-depth explanations of e.g. LSTMs
MM
I'm so glad to take this course and build my knowledge regarding time-series data and modern approaches to create prognostic models. Thanks to Andrew Ng and L. Moroney to provide this course.
SP
Few hands on programming assignments could be better for experience as was the case with starting two courses. Overall good course and the structure was well laid. Thanks for building it up
SW
Great course for review and hands-on practice of RNN/LSTM/DNN after AI specialization (Sequence Models). However, without background of DL (sequence models) it's a bit hard to catch up.
MI
I really enjoyed this course, especially because it combines all different components (DNN, CONV-NET, and RNN) together in one application. I look forward to taking more courses from deeplearning.ai.
OR
It was an amazing experience to learn from such great experts in the field and get a complete understanding of all the concepts involved and also get thorough understanding of the programming skills.
GS
Great introduction to Time series forecasting using neural networks. I liked the journey from statistics-based approaces to DL. And the application to sunspots was very interesting
WG
Though I feel some aspects of this course did not delve deep enough into the explanations of some functions, the course helped me understand how to use models for time series problems.
Showing: 20 of 801
The first week has some interesting discussion of time series data and some traditional non-ML methods for forecasting, but beyond that the course quickly divulges the all too familiar weaknesses of this specialization; lack of depth, elementary discussion, weak insight into common problems that arise during training models, and extremely poorly written quizzes that don't test the learner's gain of knowledge or skills in any meaningful way.
My biggest complaint to the instructors and the team is that for months this specialization promised the last course will discuss the WaveNet model, but the course didn't even do a cursory survey of it (In week 4, the instructor adds a Conv1D layer but doesn't even discuss the causal padding and completely skips dilations, etc, so that in effect there isn't even discussion of a single layer from WaveNet model). Sigh !
Finally, wasted my weekend and 40 euros to finish this shitty specialization. I really dont know the target audience of this specialization. If you have no background of deep learning, going through some code snippets without any explanation wont help you at all. you can't know anything behind it. If you already have some knowledge, you will find nothing new and more in this course. 1) The materials are so shallow and without any depth, just reading the slides and codes with errors. Only some high-level keras APIs are covered. The official tensorflow tutorial is much better. 2) The test questions are of no value at all, it cant test any your understanding whether about deep learning or the tool tensorflow. The assignments are poorly designed, the answers contains errors. 3) I strongly doubt the instructor, I think he does not have much ML experience. Please don't waste your money and time on this specialization. If you want to learn deep learning, go to cs230; cs231n for computer vision; cs224n and cs224u for NLP; cs20 for Tensorflow.
Very weak course, shallow, lacks content. Can be "learned" in a few hours, not weeks. Really hoped to see a working ML model for a time sequence, but the examples shown in this course do not demonstrate why bother with ML. If these examples were middle-school home work, they would be graded D+(keep trying or better use other methods). The instructor doesn't come across as an experienced ML practitioner.
Very superficial presentation of the material, and disappointing content given all the initial hype. Whatever happened to working with WaveNet? The 4 weeks to complete the course is a massive over-estimate. Expect to spend not more than a day going through the course. Quiz questions are very low value and do not test any understanding.
Unfortunately, These whole Specialization didnt match my expectations. I finished whole Deep Learning Specialization and I LOVED IT. Before starting this one I had very good feeling about this specialization; however I learned very little. Most of the videos are like "this code does this and this code does this and this line does this and this function does this etc. " . A bit disappointed, but still learned some.
I wanted to like this specialization, but I just cannot. My expectation was that this specialization would complement Andrew Ng's excellent Deep Learning specialization, but it does not. Whereas the DL specialization taught you best practices and a systematic approach to improving models, this specialization throws all of that out the window. The architectures are downright silly in some of the examples. If you want to learn TensorFlow, you would spend your time more wisely by working through the official TF tutorials, which are pretty good.
No concrete knowledge, no solid explanation. Just some demo.
I just finished the entire course on sequence learning within the trial period. I feel it was a waste of time: Hardly any new content, nothing that inspired me really beyond what I currently do. The quality of presentation IMHO was sub-par and of low-quality content. Laurence constantly stated how the MAE went down when it actually did not significantly at all. After a while, it sounded more like that he tried to influence the observer to believe that his model tweaks had a positive effect on the prediction accuracy when in reality it had hardly any effect at all. Also, at times I got the impression that Laurence did not really fully understand what he was doing, such as pushing up and down batch sizes, learning rates, flipping layer types back and forth. It did not appear to me that there was any thought process behind or logic applied to why he was doing what he was doing. Disappointed!!!
Quite disappointed about this sequence after the awesome other 3 courses taught by Andrew Ng.
Very bad course.
This course only explains time series at a very superficial level. The coding is done through Keras. The instructor does not explain the back-end. The course does not teach us about models like ARIMAX, ARIMA etc. Just some simple problems are solved in this course. Also, in the first three weeks, toy dataset was being used. The audio is also very low. There is no graded programming assignment. The quiz questions are also below standard.
I can't figure out the target audience of this course. The documentations and various blogs, books available in the internet has more content than this course, which can be completed in a few hours.
poorly designed course, easy material, lack of depth, shallow quizzes, lab exercises on colab, lab exercises is not reflective of current tensorflow version.
It's a bit of an insult to call this thing a "course". The entire video material is maybe a bit over 1 hour for the entire "4 weeks". The quizzes (1 quiz per week) are ridiculously easy, and the (ungraded) exercises are basically - "do what we did in the lecture, only from memory".
The material itself is good, but doesn't go in depth. They introduce Huber loss, and then tell you to go read about it in Wikipedia.
Overall - low quality. Would have been great as a first week in a real "mini course" (DeepLearning specialization style), or one lecture, in a real academic course.
Disappointed.
A lot of repetition of the same methods, no clear indication on how exactly to tune the chosen NNs (for instance, how to select their order, how to tune optimzers' parameters, etc) + extremely simple quizes. In general, it looks like this whole specialisation was designed just to earn some money on a existing deeplearning.ai brand. Huge disappointment.
This course quality is so poor. Didn't understand in details. Are you kidding by explaining all of these topics in just 1 min long videos ! I won't recommend this specialization course at all.
Not much to learn..very basic course without any explaination.. neither basics are cleared nor details are discussed. i didnt get idea of having such a course online and that too on coursera!
No graded exercises at the end of the practicals. Some of the quiz questions seems to be based more around general python and in 1 situation around the presenters only thoughts. Some information about estimating optimal learning rates was incorrect and misleading
It was an amazing experience to learn from such great experts in the field and get a complete understanding of all the concepts involved and also get thorough understanding of the programming skills.
Super repetitive, same code is shown like in 5 videos, IMO not the right things are emphasized (e.g. it is mentioned in every video that you should use Tensorflow v2, but some new TensorFlow commands that you come acrosse are not even mentioned with a single word). The performance differences between different types of networks do not become apparent. Too mich time "wasted" on synthetic timeseries generation and non-deep-learning (statistical) analysis. No real hands-on (letting students copy-paste code that you have just seen in the lecture is a joke!)
Was very short and summarised and it mainly gives you a model and tells you to trial and error with the hyperparameters till something clicks. very disappointing
Ok, this course was amazing, cause i pass a big large course in Udemy about Data Science for get a right way to complete my master degree tesis, and it was not enough for my, this course will help me to use my own data set that have been streamed for some sensors to analysed and predict them, before this course i don't know that CNN and LSTM is a right way to work with time series but, nowadays i know that is a good way, congrats Laurence and Andrew.
A step by step explanation of how to use TensorFlow 2.0 for building a Neural network for sequences and time series. With detailed examples of code and of how to choose hyper-parameters.