-i coudntl even get the tensorflowjs_converter tool to work. I have a newbie Q regarding C1_W3_Assignment, section:
Run the TensorFlow.js Converter on The Saved Keras Model
What is this cell in this coursera webpage exactly? Is it like a windows PC command line? Or is it python IDE? and NOT an OS commandline. If it is Python Ide, i would need to use something like subprocess.run() to run this commandline utility from within a python IDE. What is the expectation here? Also, is there new libraries that must be loaded (import os) or pip packages ( pip install pyyaml h5py) that must be installed in order for this to run?
I can save the model, but have tried many times with variations of the following syntax. But it will not run. I think that the version on this coursera webapge may be different to the current keras docs. Whats wrong with this?
tensorflowjs_converter --input_format keras --output_format tfjs_layers_model saved_model_path ./model_js
Also, the video in week3 (converting model to javascript) does extra stuff like generate a subdir using a timestamp. But this is not necessary or even relevant for this exercise, right? And another differernce, the function used for saving the model has 2 parameters:
tf.keras.experimental.export_savedmodel(model,saved_model_path)
Again, this is NOT relevant to the exercise right?
and finally finally… This ‘=’ when to use it and when not to use it in the command?
is this correct ?
–input_format=keras_saved_model
or is this correct?
–input_format keras_saved_model
i saw one thing, by mistake after watching it a few times, that change the execution behaviour. In the video he executed with ‘!’ in front of the command (!)
so once i tried !tensorflowjs_converter --input_format=keras_saved_model
etc etc … i got different errors -
2024-12-10 10:12:44.379374: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libcuda.so.1’; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory
Why doesnt the example have the name of the input model? (ie mymodel.h5 or soemthing like that) And why would it create a subfolder with the ‘unix epoch time’. Other thatn being a way to create a unique subfolder, does this unix epoch time value have any other significance?
thanks for the response Deepti. i cant even get this tool to work on a windows PC from the command line , with absolute directory addresses. If i cant get it working in this simple env, then i cant get it working in the unknown coursera jupyter env. I have downloaded the model. There are piles of warning about numpy versions and deprecated stuff, but only 2 things that look like an actual error. Howver no file is created. How long would it take to create this file? Could it take 1 hour for example?
on my windows 10 pc command prompt (administrator elevation) , the following supersimple line does not work : tensorflowjs_converter --input_format=keras_saved_model c:\week3\my_model.h5 c:\week3\my_json_model
main errors seeem to be…
File “C:\Python312\Lib\site-packages\tensorflowjs\converters\keras_h5_conversion.py”, line 33, in
from tensorflowjs import write_weights # pylint: disable=import-error
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File “C:\Python312\Lib\site-packages\tensorflowjs\write_weights.py”, line 25, in
from tensorflowjs import read_weights
File “C:\Python312\Lib\site-packages\tensorflowjs\read_weights.py”, line 28, in
np.uint8, np.uint16, np.object, np.bool]
what does success look like? It turns out that a folder was created with shard files. Its just that i only saw this hours later after some form of accidental refresh. Ive no idea what code created it so i will delete it and rerun the latest code line again…
it looks like there is no way to delete this folder…is there a way to delete this folder?