I was using TensorFlow 1.13 and Keras for my research projects. Nowadays, due to some future warnings, I installed TensorFlow 2.0 and tried to use it.
Instead of using Keras as I did before, I used tf.keras and built the same RNN model. i.e.
from keras.layers import Dense (I used before)
v.s.
from tf.keras.layers import Dense (I tried now)
All other codes are the same. However, I get some worse results for using import from tf.keras.layers one. And I am pretty sure it's not a coincidence, I tried cross-validation and run the models many times.
Does anyone have some ideas about why it happens? Are there any differences from the tf.keras.layers and keras.layers? If so, how can we be careful in case we got some "wrose" results?
tf.keras is tensorflow's implementation of the keras api. Ideally, using tf.keras should not provide you worse results. However, there might be a mismatch in the versions of both the keras which may/may not give you different results. You can check the version using tf.keras.version function and see if that is the same version of keras that you had used before.
For more details refer:
https://www.tensorflow.org/guide/keras/overview
Related
import tensorflow as tf
from tensorflow import keras
from keras import backend as K
What is the reason behind using the command—>
from keras import backend as K
What does it do? I would appreciate it if anyone explains it the simple way so that it does not get complicated in the mind.
You can find more information on what Keras backend actually is here or here.
In simpler terms to understand what Keras backend actually is
Keras is a model-level library that provides high-level building blocks for developing deep learning models. Keras does not provide low-level operations such as tensor multiplication and convolution. Instead, it relies on a specialized, well-optimized tensor library that serves as Keras' "backend engine". Instead of choosing one single tensor library and tying your Keras implementation to that library, Keras handles the problem in a modular way, allowing you to seamlessly connect multiple different backend engines to Keras.
Keras backend will allow you to write custom code or in a particular case a new "Keras module" for your use case that can support Theano and/or Tensorflow both. Like instead of tf.placeholder() you could write keras.backend.placeholder() which will work across both the libraries mentioned earlier.
I've trained a model on tensorflow 2.3 and now am trying inference on tf2.6. The model was originally trained on AWS so the CUDA and CUDNN versions are different, as well as the rest of the environment (python 3.7.11 and 3.8.12, numpy 1.18.5 and numpy 1.19.5).
The model is the exact same, the weights are also the same. However the resulting inference of the same input data results in two different vectors. They are NOT the same, differing substantially.
What is causing this issue and how do I make the results the same?
Not a solution but found a workaround. It seems to be keras 2.6.0 being problematic. Downgrading from tf2.6 to tf2.5 makes it fine as the results are the same.
I have a CNN that trains, on a few hundred thousand examples, to a validation accuracy of ~95% after one epoch. It's straight forward code, using Keras to define a network using the Sequential API. Originally I prepared and used this model on TF 1.3. When I port it over to TF 2.1, replacing the keras calls with tensorflow.keras, it gets to ~60% quickly and gets stuck there (seemingly for many epochs), and the training loss always seems to converge to the same value.
If I add in tf.disable_v2_behavior() at the top of the script, it trains similarly to before.
The documentation states simply that "It switches all global behaviors that are different between TensorFlow 1.x and 2.x to behave as intended for 1.x". Hidden behind the Keras API, I haven't found a clear answer to what this really means in practice. Why should I expect a VGG-like CNN, defined using Keras and trained with model.fit(), to work well without v2 behaviour but to fail so consistently with?
Edit: disable_eager_execution() produces the same result, with improved performance.
Please try disabling eager execution and see if that helps.
tf.compat.v1.disable_eager_execution()
(Add this to the top of your script)
I'm learning TensorFlow and Keras. I'd like to try https://www.amazon.com/Deep-Learning-Python-Francois-Chollet/dp/1617294438/, and it seems to be written in Keras.
Would it be fairly straightforward to convert code to tf.keras?
I'm not more interested in the portability of the code, rather than the true difference between the two.
The difference between tf.keras and keras is the Tensorflow specific enhancement to the framework.
keras is an API specification that describes how a Deep Learning framework should implement certain part, related to the model definition and training.
Is framework agnostic and supports different backends (Theano, Tensorflow, ...)
tf.keras is the Tensorflow specific implementation of the Keras API specification. It adds the framework the support for many Tensorflow specific features like: perfect support for tf.data.Dataset as input objects, support for eager execution, ...
In Tensorflow 2.0 tf.keras will be the default and I highly recommend to start working using tf.keras
At this point tensorflow has pretty much entirely adopted the keras API and for a good reason - it's simple, easy to use and easy to learn, whereas "pure" tensorflow comes with a lot of boilerplate code. And yes, you can use tf.keras without any issues, though you might have to re-work your imports in the code. For instance
from keras.layers.pooling import MaxPooling2D
Would turn into:
from tensorflow.keras.layers import MaxPooling2D
The history of Keras Vs tf.keras is long and twisted.
Keras: Keras is a high-level (easy to use) API, built by Google AI Developer/Researcher, Francois Chollet. Written in Python and capable of running on top of backend engines like TensorFlow, CNTK, or Theano.
TensorFlow: A library, also developed by Google, for the Deep Learning developer Community, for making deep learning applications accessible and usable to public. Open Sourced and available on GitHub.
With the release of Keras v1.1.0, Tensorflow was made default backend engine. That meant: if you installed Keras on your system, you were also installing TensorFlow.
Later, with TensorFlow v1.10.0, for the first time tf.keras submodule was introduced in Tensorflow. The first step in integrating Keras within TensorFlow
With the release of Keras 2.3.0,
first release of Keras in sync with tf.keras
Last major release to support other multi-backend engines
And most importantly, going forward, recommend switching the code from keras to Tensorflow2.0 and tf.keras packages.
Refer this tweet from François Chollet to use tf.keras.
That means,
Change Everywhere
From
from keras.models import Sequential
from keras.models import load_model
To
from tensorflow.keras.models import Sequential
from tensorflow.keras.models import load_model
And In requirements.txt,
tensorflow==2.3.0
*Disclaimer: it might give conflicts if you were using an older version of Keras. Do pip uninstall keras in that case.
As Keras becomes an API for TensorFlow, there are lots of old versions of Keras code, such as https://github.com/keiserlab/keras-neural-graph-fingerprint/blob/master/examples.py
from keras import models
With the current version of TensorFlow, do we need to change every Keras code as?
from tensorflow.keras import models
You are mixing things up:
Keras (https://keras.io/) is a library independent from TensorFlow, which specifies a high-level API for building and training neural networks and is capable of using one of multiple backends (among which, TensorFlow) for low-level tensor computation.
tf.keras (https://www.tensorflow.org/guide/keras) implements the Keras API specification within TensorFlow. In addition, the tf.keras API is optimized to work well with other TensorFlow modules: you can pass a tf.data Dataset to the .fit() method of a tf.keras model, for instance, or convert a tf.keras model to a TensorFlow estimator with tf.keras.estimator.model_to_estimator. Currently, the tf.keras API is the high-level API to look for when building models within TensorFlow, and the integration with other TensorFlow features will continue in the future.
So to answer your question: no, you don't need to convert Keras code to tf.keras code. Keras code uses the Keras library, potentially even runs on top of a different backend than TensorFlow, and will continue to work just fine in the future. Even more, it's important to not just mix up Keras and tf.keras objects within the same script, since this might produce incompatabilities, as you can see for example in this question.
Update: Keras will be abandoned in favor of tf.keras: https://twitter.com/fchollet/status/1174019423541157888