Ggtcf

Is it canonical bit space?

Can one be a co-translator of a book, if he does not know the language that the book is translated into?

Twin primes whose sum is a cube

AES: Why is it a good practice to use only the first 16bytes of a hash for encryption?

Is "remove commented out code" correct English?

Can a virus destroy the BIOS of a modern computer?

Arrow those variables!

Is it legal for company to use my work email to pretend I still work there?

In Romance of the Three Kingdoms why do people still use bamboo sticks when paper had already been invented?

Should I tell management that I intend to leave due to bad software development practices?

What exploit are these user agents trying to use?

How much of data wrangling is a data scientist's job?

Facing a paradox: Earnshaw's theorem in one dimension

Why is it a bad idea to hire a hitman to eliminate most corrupt politicians?

Is it possible to create light that imparts a greater proportion of its energy as momentum rather than heat?

Why doesn't H₄O²⁺ exist?

Can I ask the recruiters in my resume to put the reason why I am rejected?

I would say: "You are another teacher", but she is a woman and I am a man

Why can't we play rap on piano?

Brothers & sisters

Emailing HOD to enhance faculty application

What reasons are there for a Capitalist to oppose a 100% inheritance tax?

Why "Having chlorophyll without photosynthesis is actually very dangerous" and "like living with a bomb"?

How to prevent "they're falling in love" trope

Proper conversion of contrib.HashTable usage in TF 2

Usage of __slots__?Proper way to use **kwargs in PythonProper way to declare custom exceptions in modern Python?Proper indentation for Python multiline stringsTensorflow and Multiprocessing: Passing SessionsReplicating models in Keras and Tensorflow for a multi-threaded settingValueError("Cannot evaluate tensor using `eval()`:Tensorflow feature columns in Dataset map Table already initialized issueHow to get explicit session for input_fn when training a TensorFlow modelHow to call a Keras function from tensorflow Graph for multi-task learning?

.everyoneloves__top-leaderboard:empty,.everyoneloves__mid-leaderboard:empty,.everyoneloves__bot-mid-leaderboard:empty height:90px;width:728px;box-sizing:border-box;

0

I have a Keras layer that takes a string input and returns a series of indices into an embedding matrix. The code for testing the TF 1.x version is here, and the theoretically-TF2-compatible testing code is here.

The TF2 version doesn't work, and it seems that my issues center around initializing the table. In the old version I need to run the tables_initializer (although I believe Keras.fit does this), but this isn't really something that you can do in TF2 -- both because there's no table_initializer and also because there's no access (afaict) to explicit Sessions.

So when I run the TF 2 version of this, I need to figure out how to run the initializer on the call to keras.layer.build. The most obvious way to do this is to follow the documentation which directs StaticHashTable.init.run() -- this is completely incorrect however.

So if I don't do any of this (on the assumption that it will somehow magically work), I get a tensorflow.python.framework.errors_impl.FailedPreconditionError: Table not initialized.

So, an alternative: init just returns the initializer, so I could instead take the table's initializer and do: KeyValueInitializer.initialize(StatisHashTable).run(), but this still requires a default Session to be registered, which, even if possible, does not seem like idiomatic TF 2 (and I haven't figured out how to get a 'Session' anyways). The specific error I get is ValueError: Cannot execute operation using 'run()': No default session is registered. Use 'with sess.as_default():' or pass an explicit session to 'run(session=sess)'.

So that's a dead end. I've tried some other stuff that I'm fairly certain are not 'how you are supposed to do it' in TF 2 like wrapping the initializer Operator in control_dependencies or giving that its run method a session from compat.v1.keras.backend.get_session() all to no avail.

I'd like to believe that the tf.lookup.StaticHashTable works somehow in TF 2, but I'm just not sure how to make it work -- anyone have any advice?

python tensorflow

share|improve this question

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

add a comment | 

0

I have a Keras layer that takes a string input and returns a series of indices into an embedding matrix. The code for testing the TF 1.x version is here, and the theoretically-TF2-compatible testing code is here.

The TF2 version doesn't work, and it seems that my issues center around initializing the table. In the old version I need to run the tables_initializer (although I believe Keras.fit does this), but this isn't really something that you can do in TF2 -- both because there's no table_initializer and also because there's no access (afaict) to explicit Sessions.

So when I run the TF 2 version of this, I need to figure out how to run the initializer on the call to keras.layer.build. The most obvious way to do this is to follow the documentation which directs StaticHashTable.init.run() -- this is completely incorrect however.

So if I don't do any of this (on the assumption that it will somehow magically work), I get a tensorflow.python.framework.errors_impl.FailedPreconditionError: Table not initialized.

So, an alternative: init just returns the initializer, so I could instead take the table's initializer and do: KeyValueInitializer.initialize(StatisHashTable).run(), but this still requires a default Session to be registered, which, even if possible, does not seem like idiomatic TF 2 (and I haven't figured out how to get a 'Session' anyways). The specific error I get is ValueError: Cannot execute operation using 'run()': No default session is registered. Use 'with sess.as_default():' or pass an explicit session to 'run(session=sess)'.

So that's a dead end. I've tried some other stuff that I'm fairly certain are not 'how you are supposed to do it' in TF 2 like wrapping the initializer Operator in control_dependencies or giving that its run method a session from compat.v1.keras.backend.get_session() all to no avail.

I'd like to believe that the tf.lookup.StaticHashTable works somehow in TF 2, but I'm just not sure how to make it work -- anyone have any advice?

python tensorflow

share|improve this question

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

add a comment | 

I have a Keras layer that takes a string input and returns a series of indices into an embedding matrix. The code for testing the TF 1.x version is here, and the theoretically-TF2-compatible testing code is here.

The TF2 version doesn't work, and it seems that my issues center around initializing the table. In the old version I need to run the tables_initializer (although I believe Keras.fit does this), but this isn't really something that you can do in TF2 -- both because there's no table_initializer and also because there's no access (afaict) to explicit Sessions.

So when I run the TF 2 version of this, I need to figure out how to run the initializer on the call to keras.layer.build. The most obvious way to do this is to follow the documentation which directs StaticHashTable.init.run() -- this is completely incorrect however.

So if I don't do any of this (on the assumption that it will somehow magically work), I get a tensorflow.python.framework.errors_impl.FailedPreconditionError: Table not initialized.

So, an alternative: init just returns the initializer, so I could instead take the table's initializer and do: KeyValueInitializer.initialize(StatisHashTable).run(), but this still requires a default Session to be registered, which, even if possible, does not seem like idiomatic TF 2 (and I haven't figured out how to get a 'Session' anyways). The specific error I get is ValueError: Cannot execute operation using 'run()': No default session is registered. Use 'with sess.as_default():' or pass an explicit session to 'run(session=sess)'.

So that's a dead end. I've tried some other stuff that I'm fairly certain are not 'how you are supposed to do it' in TF 2 like wrapping the initializer Operator in control_dependencies or giving that its run method a session from compat.v1.keras.backend.get_session() all to no avail.

I'd like to believe that the tf.lookup.StaticHashTable works somehow in TF 2, but I'm just not sure how to make it work -- anyone have any advice?

python tensorflow

share|improve this question

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

share|improve this question

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

share|improve this question

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

asked Mar 9 at 0:00

zmjjmzzmjjmz

135

asked Mar 9 at 0:00

zmjjmzzmjjmz

135