Deploying Machine Learning Models Online with Watson Machine Learning | Python Scikit-Learn
ฝัง
- เผยแพร่เมื่อ 18 ธ.ค. 2020
- You’ve scoured the web and collected a ton of data.
You’ve done a bunch of Exploratory Data Analysis.
You’ve trained a bunch of models.
They’re pretty accurate.
Now what?
Now, you deploy them! This is one of the final steps in the CRISP-DM lifecycle. Once you’ve evaluated your models and you’re happy to release them into the wild, this is generally the step you’re going to want to take.
But how?
Well, in this video you’ll learn exactly how to do that. You’ll learn step by step how to take a trained Scikit-Learn model and deploy it to the Cloud using Watson Machine Learning. This model can then be used for a whole bunch of applications and can even be used in different languages like Javascript, Scala, Java and Go. And what’s best? You’ll do it all in less than half an hour.
In this video you’ll learn how to:
1. Saving models to Watson Machine Learning
2. Creating online deployments with Python
3. Scoring your model using the Python API
GET THE CODE!
github.com/nicknochnack/Deplo...
Links Mentioned
IBM Cloud: cloud.ibm.com/login
Watson Machine Learning: cloud.ibm.com/catalog/service...
Watson Studio: cloud.ibm.com/catalog/service...
Example deploying with Cloud Pak for Data: github.com/IBM/watson-machine...
Watson Machine Learning Endpoint URLs
* Dallas: us-south.ml.cloud.ibm.com
* London - eu-gb.ml.cloud.ibm.com
* Frankfurt - eu-de.ml.cloud.ibm.com
* Tokyo - jp-tok.ml.cloud.ibm.com
If you have any questions, please drop a comment below!
Oh, and don't forget to connect with me!
LinkedIn: / nicholasrenotte
Facebook: / nickrenotte
GitHub: github.com/nicknochnack
Happy coding!
Nick
P.s. Let me know how you go and drop a comment if you need a hand! - วิทยาศาสตร์และเทคโนโลยี
you're the best one on youtube, no doubts!
congrats for everything and thank you!
Honestly some of the best ml content I’ve seen in a long time, keep it up!
Woah, thanks so much! Definitely a ton more planned @4K!
thanks so much for the regular uploads.
😀 trying my best for two a week...plennnnty more to come @John!
A great content and amazing explanation. Thanks for sharing this.
Great content as always
Thanks for your contents, really practical and the way you explain it is very easy to understand!
Thanks so much @Harry, glad you enjoyed it!
Does the deployment part work the same if we trained our model in jupyter(ex. action recognition video) we save it into .h5 because I tried to import it to Watson IBM but didn't work. I actually want to do explainability AI
can I deploy my inception v3 model on IBM cloud free tier ?? won't it take too much space. Please help??
best video
I've been trying it for a week, is Watson unable to deploy the Sklearn model anymore?
Hi, I can't access IBM, is there any alternatives to IBM and do you have any resources on it? Thanks
wml_client = WatsonMachineLearningAPIClient(wml_credentials)
in this line it is saying key error : instance_id
Nice!
🙏🙏 Thanks so much @Gustavo!
im tryin to create any account with ibm but they werent able to verify my credit card info and i cant find a number to call. any tips for this issue?
3 vids in 1 week! Love the tutorials. Also wondering if you can make something on training your own image inpainting model, there seems to be a lack of resources for that online
Hmmm looks sick @Andy, let me take a look. So it's sort of like image restoration?
@@NicholasRenotte Yeah, I kinda want to try training a model on my own dataset so you can restore specific types of images or do watermark removal.
@@Andy-rq6rq hmmm that's an awesome use case. Will definitely take a look. I have some leave coming up so I'm going to start prototyping some new stuff for TH-cam!
@@NicholasRenotte Awesome! Thanks for considering it
Looks sick @@Andy-rq6rq, if you've got any other ideas shoot 'em through!
Hi thanks for your great videos, do you have any tutorial deploying watson models to android studio?
result=wml_client.deployments.score(deployment_uid, payload); result
getting "ApiRequestFailure: Failure during scoring." error
"message": "Invalid request entity: Unsupported software specification 'default_py3.7'." how to resolve this error , plz someone give solution to this
i tried in so many ways and i cant figure out a solution
🔥🔥🔥This guy is on fire!
😅 thanks @Vikash!! What did you think of the little explainer cards e.g. 04:47? Yay or nay?
It's nice bro, makes it easier for us to make personal notes 🔥
@@vikashchand. awesome, so it stays!!
@@NicholasRenotte
th-cam.com/video/dDIk1Tmnj9A/w-d-xo.html 🔥🤘
@@vikashchand. oooooh yeah this is sick, on the list!
When I move to 23:41, "fields":X_test.columns.to_numpy().tolist() shows " 'Index' object has no attribute 'to_numpy' ", why??
Can you try passing X.reset_index(inplace=True). Looks like the df might be indexed!
I dont have credit card so how can i create ibm account? Someone helps me pls
When I move to 21:34, my deployment is failed. it shows that "No module named 'sklearn.externals.joblib". I google it a lot. it seems that in scikit-learn_0.23, the joblib is deleted. Do you have some idea to solve this problem??
Heya, can you share your code? Are you explicitly using Joblib?
@@NicholasRenotte drive.google.com/drive/folders/16TRX9R-Qqktpmr8uOHgS6N75QaFwuMC8?usp=sharing
@@NicholasRenotte can you download my code??
@@frankdong7396 yup, let me take a look!
Can I use watson to train the moedl? My laptop can't handle it
Yup, check this out: th-cam.com/video/rr-KwIjinpM/w-d-xo.html
I need help to deploy python function. After deployment it throws an error mentioning that no module named project_lib found ... I used this library for reading data from database connection is inside the function that i deployed. While deployment i used default_python_opence as software specification. Can u please help me to resolve this error? Please post any video for reference regarding python function deployment in IBM watson studio.
Presumably you're using a custom function right? In certain circumstances, you'll need to deploy and install new libs into the deployment. Check this out: dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/install-cust-lib.html
@@NicholasRenotte Thank you much for the reply. I tried installing the dependent library(project_lib) inside the function. I thought after deployment, it will be installed in the runtime environment, whenever the function is called. But the thing is after deployment while scoring now it shows an error like get_iPython is not defined... when I searched for this error it says that get_Ipython is called while using pip install command. I tried to install project_lib inside the function, but now it throws an error in installing it itself as it requires get_Ipython to install libraries. I don't know how to solve it. Need your help, please.
@@sundarrajr220 do you need the project_lib to deploy your model?
Thanks for simplifying the concepts. I'm stuck at wml_client creation. It's not showing any space list. Please help
Heya @Sandeep, you need to create a space inside of dataplatform.cloud.ibm.com/ first. Shown here: th-cam.com/video/tUHl5TiP_oA/w-d-xo.html
@@NicholasRenotte thanks for immediate response. It solved my issue. Thanks again. 😊
Dear@@NicholasRenotte, thanks for all the immediate support and suggestionsm
I've deployed the model onto the cloud. Now may I know what is next? I mean, how can we use this deployed model predict the new data?
Hi @@sandeepmandrawadkar9133, this is shown here: th-cam.com/video/tUHl5TiP_oA/w-d-xo.html
Dear@@NicholasRenotte thanks a lot for all your support 🙏
Hi, I am a big fan of you. BTW, what is Buckets and Buckets at 8:39, and when did you create Machine Learning -9g. It is my first time doing it and my screen shows nothing in those two parts. Big Thank
Heya @Xingcheng, buckets and buckets is the cloud object store service (cloud.ibm.com/objectstorage/create), if you create a free tier of the service you should be able to select and associate it!
@@NicholasRenotte Thanks for your rely. That part is working. But a new problem occurred. When I move to 21:34, my deployment is failed. it shows that "No module named 'sklearn.externals.joblib". I google it a lot. it seems that in scikit-learn_0.23, the joblib is deleted. Do you have some idea to solve this problem??
2021-01-05 17:50:36,629 - ibm_watson_machine_learning.wml_client_error - WARNING - Deployment creation failed for deployment id: 1794791a-05c6-4b7b-aab4-d807bdadec5d. Errors: [{'code': 'invalid_model_archive', 'message': "Model loading through joblib failed with exception: ModuleNotFoundError and message: No module named 'sklearn.externals.joblib'. The model object provided for deployment is either invalid or trained using a version of scikit-learn other than: 0.23", 'target': {'type': 'none', 'name': 'none'}, 'more_info': 'none'}]
I am out of solutions. do we have any other software to replace ibm-watson-machine-learning, or can we write code on ibm-watson-machine-learning?? Stucked so much......
@@frankdong7396 hmmm, interesting, that's good to know. I did a little digging looks like it should work by downgrading to 0.22.2. Try running this:
!pip install -U scikit-learn==0.22.2
Superb! , but sir when I tried list down the deployment space , it displayed none , but I did create the deployment space , so would u mind giving me some suggestions?
what should I do with it sir, looking forward to hearing back from you? Thanks
Hmmm, weird, you're already authenticated and have an active account?
nice and great content ! this is what im looking for !
if i may, can you make watson openscale for machine learning ?
Ooooh yeah, actually present on it over the next couple of weeks at a few conferences!
@@NicholasRenotte what conferences? Ibm think? Hit me the link or invitation please
@@kevinelfri8460 next week at the AWS Summit: aws.amazon.com/events/summits/online/anz/ I'm on the 18th@2:45 and 19th@2ish!
Instead of retraining my model and deploying it again, is there any possibility to deploy my re-trained and fine-tuned model in the same deployment id? My requirement is From the front end screen they will click an ok button or reject button (our deployed model will be used as an API call in the script when they click the button). If they click the ok button then our deployed model should be called and prediction results will be received to the frontend as a response. And when they click the reject button we need to retrain the model by hyper parameter tuning and then with that model we need to predict results, inorder to show them changes in the predictions between clicking the ok and reject button.
I have an idea like we can schedule a job monthly once so that the model will be retrained accordingly and stored in the same deployment id itself because they will be using this portal only once in a month that too on a specific date... any suggestions
Sure can, that's industry standard process with WML. You're able to redeploy over the same endpoint.
@@NicholasRenotte i searched a lot to redeploy over the same endpoint but i couldn't find any relative blog for the same... can u please share me any referral link
@@sundarrajr220 hmmm, I don't think I've got any public doco just my random code unfortunately.
Hello Nicholas, I'm bethel I love ur tutorials. I just completed the IBM Data Science professional certificate. I need u to mentor me. Would it be possible for u to do so, so I can get more experience and learn more about the field.
Why there are just 4 regions ? , Is tutorial irrelevant for me if I belong to another region ? I am relly misunderstand why we need region for deploy ML model...
Heya @Peter, this just refers to where the model will be hosted, tbh it doesn't really matter apart from minimizing load times. I'd just pick the region that's closest to you!
@@NicholasRenotte Thanks for clear answer :)
@@petertomasovic21 anytime! Let me know if you need anything else!
Very legible content, though I am facing a issue which says unsupported software specification: "default_py3.7"
Would you mind pls tell me how to resolve this
Hi, what a amazing Machine Learning content, I am pretty glad with all your videos I've seen.
With this one, I'm having the next error when I type this code line:
from ibm_watson_machine_learning import APIClient
import json
import numpy as np
then appears --> ContextualVersionConflict: (docutils 0.16 (c:\users\gabri\anaconda3\lib\site-packages), Requirement.parse('docutils=0.10'), {'ibm-cos-sdk-core'})
Again, THANK YOU! it's amazing all you do.
Actually, the error is in:
from ibm_watson_machine_learning import APIClient
I think is some issue with the version but I don't know how to fix it
Ha, weird, is the error the same @Nuestro? Just this?
ContextualVersionConflict: (docutils 0.16 (c:\users\gabri\anaconda3\lib\site-packages), Requirement.parse('docutils=0.10'), {'ibm-cos-sdk-core'})
@@NicholasRenotte yes, it's really weird and I had the feeling that is about the version of Python. I'm working with Python 3.8 and in some forums says that IBM works with Python 3.7. Which version are you working with?
Thanks again, great job all your tutorials.
@@nuestromundoxl9871 yep, I'm working with 3.7.3. Possibly try that?
@@NicholasRenotte Yeess, I'm going to try.
Hi, I made everything work!!!!! Oh, God. The problem is the version of sklearn and python. We should use python 3.6 rather than 3.7, and sklearn 0.22 rather than 0.23. Just as the links shows: www.ibm.com/support/knowledgecenter/SSQNUZ_2.5.0/wsj/wmls/wmls-deploy-python.html
So you got it working!?! EXCITEDDD!
@@NicholasRenotte Yeah, everything working. Thanks a lot....
@@frankdong7396 awesome stuff!!
TOO MUCH HAND MOVEMENTS