I like the serverless solution. I also found some concepts explained too quickly and had to rewind. A suggestion is selecting a theme e.g. for this video like "catching a train". The station is the cloud service. Platforms define which version is selected. Passengers are the web traffic. Selecting a theme for each video or allows people like me who appreciate the tech. but still learning the details connect the dots quicker and while watching because I'm familiar with trains. Well done 4.5 *'s
philiptwayne Hi, but what's the "prediction service" here? He said that we can turn a trained model into a "prediction service" in minutes, so what's exactly it? an application or function of the Cloud Machine Engine?
Please include an end-to-end tutorial as well. You are using the Experiment object to get the saved model. But this is not the way many people learn tensorflow using online materials. The tutorial should show step-by-step how a model like the one introduced by Martin Gorner can be converted to a savedModel and deployed on gcloud. This should also show the process needed for someone unfamiliar with google cloud how to expose the model via rest api (e.g. showing the oAuth or service account key auth process etc.). I had to figure it out myself. It felt great to solve such a complex problem (I am not talking about the DNN model! believe me!). But if you want to make it more popular, you should provide better end-to-end tutorials.
Why Google's videos are either too complicated/specialized or too simple? Titles also do not match what is in the video. Where is the "at scale" part in this video?
Question: Does artificial intelligence have the ability to create a similar model without a biological sample? The purpose of asking this question: The production structure is based on trial and error principles. How can the final model be completely valid from an unknown experience?
Once your model is uploaded to Cloud ML Engine, how do you get predictions back in your Iris example? I understand that you can use the CLI command: gcloud ml-engine predict --model iris_model --version v1 --json-instances sample.json. What would the structure of the sample json file be?
Yufeng, why is it that you and Josh Gordon are speaking so blazingly fast as to cram the maximum amount of teaching into 5-7 minute videos? Why not make the vids a little bit longer so you can actually take a breath? love the vids.
I love this guy
I like the serverless solution. I also found some concepts explained too quickly and had to rewind. A suggestion is selecting a theme e.g. for this video like "catching a train". The station is the cloud service. Platforms define which version is selected. Passengers are the web traffic. Selecting a theme for each video or allows people like me who appreciate the tech. but still learning the details connect the dots quicker and while watching because I'm familiar with trains. Well done 4.5 *'s
philiptwayne Hi, but what's the "prediction service" here? He said that we can turn a trained model into a "prediction service" in minutes, so what's exactly it? an application or function of the Cloud Machine Engine?
Please include an end-to-end tutorial as well.
You are using the Experiment object to get the saved model. But this is not the way many people learn tensorflow using online materials. The tutorial should show step-by-step how a model like the one introduced by Martin Gorner can be converted to a savedModel and deployed on gcloud.
This should also show the process needed for someone unfamiliar with google cloud how to expose the model via rest api (e.g. showing the oAuth or service account key auth process etc.).
I had to figure it out myself. It felt great to solve such a complex problem (I am not talking about the DNN model! believe me!). But if you want to make it more popular, you should provide better end-to-end tutorials.
Give this guy a medal !!
Why Google's videos are either too complicated/specialized or too simple? Titles also do not match what is in the video. Where is the "at scale" part in this video?
Question: Does artificial intelligence have the ability to create a similar model without a biological sample? The purpose of asking this question: The production structure is based on trial and error principles. How can the final model be completely valid from an unknown experience?
Why do you recommend a static solution? It seems that a dynamic system that served predictions while also continuing to learn would be better.
The lecture is very clear and concise! Great job!
Once your model is uploaded to Cloud ML Engine, how do you get predictions back in your Iris example? I understand that you can use the CLI command: gcloud ml-engine predict --model iris_model --version v1 --json-instances sample.json. What would the structure of the sample json file be?
cloud.google.com/ml-engine/docs/v1/predict-request
You can find it here :)
Now, we need to call the prediction service! Will be done in the next video?
What is the commercial big picture for tensorflow? What is the dream?
Why is the playlist in the wrong order?
Great job Guo!!!
Yufeng, why is it that you and Josh Gordon are speaking so blazingly fast as to cram the maximum amount of teaching into 5-7 minute videos? Why not make the vids a little bit longer so you can actually take a breath?
love the vids.
Great guy!
Wow! u finally changed your shirt ... Just kidding😂, thanks for the videos
the beginner gains is really awkward lmfao