👉 Check out the blog post and other resources for this video: 🔗 deeplizard.com/learn/video/7n1SpeudvAE 👀 Come say hey to us on OUR VLOG: 🔗 th-cam.com/users/deeplizardvlog
Mandy, your Jupyter Code runs beautifully in VSCode step-by-step with your video. Your narration speed is perfect. What more can I say, a perfect learning experience in every aspect. Deeplizard is hands down the Best of Breed in its class.
Thank you very much for this video! It was really eye-opening and this is gonna help me a lot with my assignment, especially with learning how to load models created by others!
Great content! I'm struggling to save my custom tf.keras model to a .pb file, in order to deploy it with opencv dnn. It would be awesome if you covered this topic in a future video!
Thank you for this insightful tutorial! Question: After viewing the weights, how does one find the most important inputs in the model? That is, the ones that contribute more to the output?
I think I can give the viewer more infomation about: 1. The 1st way have some problems : 1.1. Embedding weights not save 1.2 Custom function should declear both save and load 1.3 Stage only save the steps (no epochs) 2. 2 & 3 use together BTW. I dont know why . I tell some feelings. When I first saw the videos before . I can feel the passion or know something I dont know . EG. the east egg after each video makes me think . But now I found myself just go to school to listen to the teacher . Yeah , another class I have to take. just like that . I dont know what happend. Just feelings. Hope to give some feedback to your guys .
Always appreciate your thoughtful comments and feedback! This course is more instructional in nature in how to use the Keras API step-by-step. I think the Easter eggs you're referring to are the deeper insights we would share at the end of the Reinforcement Learning episodes and some of the PyTorch episodes, right? Since we're only focusing on using the use of the API in this course (so far, at least), it didn't seem as fitting to include the deeper insights here. Keep the feedback coming, always happy to hear from you! :)
@@deeplizard this come from japanese word そっか which should pronunce to sokka but sogga in daily life. It measns So it turns out . Its popular in Chinese young generations.
Good tutorial. Question, after training and saving a model, is there a way to multiply the trained weights before loading them into a new model so I can see how it performs w/o training again? for example: (trained weights)*.95
Very clean explanation, thank you! I have a doubt, hope it'd get cleared. Can't we save a Keras models under .pb extension? If we can, can u make a tutorial on that. Because a large amount of 3rd party softwares(like OpenVINO) makes use of models with .pb extension for making inferences. No one has given a clear explanation on how we can save it under .pb extension. If you can make a tutorial on that, it'll be really helpful.
In TF 2.x and later, you can use the same model.save() function but don't include ".h5" in the filename. Then I believe the Keras model will be saved as a pb file.
Hello, deeplizard! I've just trained my first model after watching your great course. I saved the model as a .pb, but I need also a .pbtxt file. I googled many ways to convert the .pb model, but none of them worked. Can you recommend a way to convert the model? The only files I got after saving the model are saved_model.pb, variables.data-00000-of-00001 variables.index.
Say we are going to save a trained CNN model. Can we save only the trained convolutional base part instead of saving the entire model including the fully connected part? The purpose is mainly to use the convolutional base as pretrained feature extractor and fine tune the fully-connected part. How can we do that?
hi, if I for instance trained a model with 3 features (for stock price prediction), close price, volume and a moving average using LSTM. The length of the sequence is e.g. the last 20 points. Then when I have saved the model and load it back up can I extract information from the saved model about which features have been used and the sequence length that is used? The Sequence length I can extract but I do not see how I can extract information about which features have been used. I guess I need to save that to a separate file? This assuming I can't remember which features I used to train that particular model. For instance if I load the model 1 week after I trained it and I do not know or can't remember the features and sequence length I used. How would you retrieve this info? Can it be retrieved from a H5 file? Thanks.
I'm working on a Dell PC with Win10 and all the latest API's. The model.save() command worked without throwing an exception but when I opened the actual .H5 file in Jupyter Notebook it said "Error!... Saving Disabled....model.h5 is not UTF encoded.....see console for more details". I googled this error for an hour or so before eventually deciding to open the file with the Windows notebook app and saw that it actually did contain lots of data not apparent through the Jupyter browser. The file also seems to faithfully reproduce the model when the load_model() function is used. So consider disregarding this error if encountered. Compliments to deeplizard for code that appears to be 100% right and scrupulously maintained.
pls, help me out. When I run the saving model cell, I get so many lines of error, and the last line is mentioned below. OSError: Unable to create file (unable to open file: name = 'models/medical_trail_model.h5', errno = 2, error message = 'No such file or directory', flags = 13, o_flags = 302) what should I do now?
ok, I figured it out..... instead of that cell..running the below lines fixed that problem.. import os.path if os.path.isfile(r'C:\Users\\medical_trail_model.h5') is False: model.save(r'C:\User\\medical_trail_model.h5')
I saved my model using model.save(path). In another python script, I loaded the model and used the model.predict() function. But costs me almost 10s. I have time constraint of 5s. Can anyone help me with this? Pls suggest me a better way to load the model and predict.
Hii everyone. AttributeError: 'str' object has no attribute 'decode' got this error while loading the saved model using load_model method. Inside the .h5 file I found the following lines. filename.h5 is not UTF-8 encoded saving disabled. see consol for more details
👉 Check out the blog post and other resources for this video:
🔗 deeplizard.com/learn/video/7n1SpeudvAE
👀 Come say hey to us on OUR VLOG:
🔗 th-cam.com/users/deeplizardvlog
Best tutorial by far in Keras... finally someone who truly teaches! Thank you a lot!
Mandy, your Jupyter Code runs beautifully in VSCode step-by-step with your video. Your narration speed is perfect. What more can I say, a perfect learning experience in every aspect. Deeplizard is hands down the Best of Breed in its class.
Thank you, Christian! 😊😊
I'm so proud of you guys :)
Just crazy! Directly-to-the-point explanation. Keep up the good work!
THANK YOU . IT WAS HELPFULL :)
Great concise tutorials! much appreciated! 👍
Oh my, this is real teaching
Thank you very much for this video! It was really eye-opening and this is gonna help me a lot with my assignment, especially with learning how to load models created by others!
Thanks for the explaination!!!
Thank you very much for this video. This topic is really very important to know.
This video has saved my ....
Thankss aa lootttt , this helped meeee
Very well explained 👏👏👏
your are awesome ❤
Thanks for your keras video playlist. Please upload one playlist on Tensorflow 2.X from beginner to advance .
Great content! I'm struggling to save my custom tf.keras model to a .pb file, in order to deploy it with opencv dnn. It would be awesome if you covered this topic in a future video!
Thank you for this insightful tutorial! Question: After viewing the weights, how does one find the most important inputs in the model? That is, the ones that contribute more to the output?
So you can use json to save model and use save weights and then combine these two so you have same function as saving model
I think I can give the viewer more infomation about:
1. The 1st way have some problems :
1.1. Embedding weights not save
1.2 Custom function should declear both save and load
1.3 Stage only save the steps (no epochs)
2. 2 & 3 use together
BTW. I dont know why . I tell some feelings. When I first saw the videos before . I can feel the passion or know something I dont know . EG. the east egg after each video makes me think . But now I found myself just go to school to listen to the teacher . Yeah , another class I have to take. just like that . I dont know what happend. Just feelings. Hope to give some feedback to your guys .
Always appreciate your thoughtful comments and feedback! This course is more instructional in nature in how to use the Keras API step-by-step. I think the Easter eggs you're referring to are the deeper insights we would share at the end of the Reinforcement Learning episodes and some of the PyTorch episodes, right?
Since we're only focusing on using the use of the API in this course (so far, at least), it didn't seem as fitting to include the deeper insights here. Keep the feedback coming, always happy to hear from you! :)
@@deeplizard soga
I don't know what soga means, and Google doesn't appear to be giving me a relevant description 😬
@@deeplizard this come from japanese word そっか which should pronunce to sokka but sogga in daily life. It measns So it turns out . Its popular in Chinese young generations.
Good tutorial. Question, after training and saving a model, is there a way to multiply the trained weights before loading them into a new model so I can see how it performs w/o training again? for example: (trained weights)*.95
Hey Mandy !! Love your explanations. I have a cliche question. What do you prefer between pytorch and tensorflow ?
Good afternoon, I have a question about retraining, how can I do retraining if I already have a trained model?
This is very good！
Please make a github that we can try much more！
good luck
Hi, can I save and load the .JSON architecture and import weights into it?
Very clean explanation, thank you!
I have a doubt, hope it'd get cleared. Can't we save a Keras models under .pb extension? If we can, can u make a tutorial on that. Because a large amount of 3rd party softwares(like OpenVINO) makes use of models with .pb extension for making inferences. No one has given a clear explanation on how we can save it under .pb extension. If you can make a tutorial on that, it'll be really helpful.
In TF 2.x and later, you can use the same model.save() function but don't include ".h5" in the filename. Then I believe the Keras model will be saved as a pb file.
@@deeplizard That worked. Thank you!
Can I save/load multiple keras neural network models into one .h5 file?
Hello, deeplizard!
I've just trained my first model after watching your great course.
I saved the model as a .pb, but I need also a .pbtxt file. I googled many ways to convert the .pb model, but none of them worked. Can you recommend a way to convert the model? The only files I got after saving the model are saved_model.pb, variables.data-00000-of-00001 variables.index.
Say we are going to save a trained CNN model. Can we save only the trained convolutional base part instead of saving the entire model including the fully connected part? The purpose is mainly to use the convolutional base as pretrained feature extractor and fine tune the fully-connected part. How can we do that?
hi, if I for instance trained a model with 3 features (for stock price prediction), close price, volume and a moving average using LSTM. The length of the sequence is e.g. the last 20 points. Then when I have saved the model and load it back up can I extract information from the saved model about which features have been used and the sequence length that is used? The Sequence length I can extract but I do not see how I can extract information about which features have been used. I guess I need to save that to a separate file? This assuming I can't remember which features I used to train that particular model. For instance if I load the model 1 week after I trained it and I do not know or can't remember the features and sequence length I used. How would you retrieve this info? Can it be retrieved from a H5 file? Thanks.
I'm working on a Dell PC with Win10 and all the latest API's. The model.save() command worked without throwing an exception but when I opened the actual .H5 file in Jupyter Notebook it said "Error!... Saving Disabled....model.h5 is not UTF encoded.....see console for more details". I googled this error for an hour or so before eventually deciding to open the file with the Windows notebook app and saw that it actually did contain lots of data not apparent through the Jupyter browser. The file also seems to faithfully reproduce the model when the load_model() function is used. So consider disregarding this error if encountered.
Compliments to deeplizard for code that appears to be 100% right and scrupulously maintained.
Good to know Garrie, thanks!
❤️❤️
does this work in google colab
Hello, I used the first method but I get an error unknown keras layer, its saying i need to pass it to the custom objects
Does it work for theano backend? I use the method but my "model.h5" didn't appear in a folder... and there is no any error message.
pls, help me out. When I run the saving model cell, I get so many lines of error, and the last line is mentioned below.
OSError: Unable to create file (unable to open file: name = 'models/medical_trail_model.h5', errno = 2, error message = 'No such file or directory', flags = 13, o_flags = 302)
what should I do now?
ok, I figured it out.....
instead of that cell..running the below lines fixed that problem..
import os.path
if os.path.isfile(r'C:\Users\\medical_trail_model.h5') is False:
model.save(r'C:\User\\medical_trail_model.h5')
@@manojrajsr6001 thanks for mention that awesome answer.
What if we have custom layers ? None of these work, might need custom objects to save?
you can pass custom objects as arguments in the load_model function and then you should be all set
please help me. When i run i have an errror:
ValueError: Unknown activation function:softmax_v2
Every where is loading but didnt say on how to use that to do prediction
I saved my model using model.save(path). In another python script, I loaded the model and used the model.predict() function. But costs me almost 10s. I have time constraint of 5s. Can anyone help me with this? Pls suggest me a better way to load the model and predict.
Hii everyone.
AttributeError: 'str' object has no attribute 'decode'
got this error while loading the saved model using load_model method.
Inside the .h5 file I found the following lines.
filename.h5 is not UTF-8 encoded saving disabled. see consol for more details
you have only this tshirt? :D
Correct 😆
i'm in love.. u so beautiful !. And good explanation, thanks!!!
simp
Plz dont make vdos in bedroom