The difference between the two is muddled in my head, notwithstanding the nuances of what is eager and what isn't. From what I gather, the #tf.function decorator has two benefits in that
it converts functions into TensorFlow graphs for performance, and
allows for a more Pythonic style of coding by interpreting many (but not all) common-place Python operations into tensor operations, e.g. if into tf.cond, etc.
From the definition of tf.py_function, it seems that it does just #2 above. Hence, why bother with tf.py_function when tf.function does the job with a performance improvement to boot and without the inability of the former to serialize?
They do indeed start to resemble each other as they are improved, so it is useful to see where they come from. Initially, the difference was that:
#tf.function turns python code into a series of TensorFlow graph nodes.
tf.py_function wraps an existing python function into a single graph node.
This means that tf.function requires your code to be relatively simple while tf.py_function can handle any python code, no matter how complex.
While this line is indeed blurring, with tf.py_function doing more interpretation and tf.function accepting lot's of complex python commands, the general rule stays the same:
If you have relatively simple logic in your python code, use tf.function.
When you use complex code, like large external libraries (e.g. connecting to a database, or loading a large external NLP package) use tf.py_function.
Related
Backgrund
In Tensorflow, even when using mutable variables, it looks there is no out option as in numpy to specify the location to store the calculation result. One of the reason why the calculation gets slower is the temporary copy as explained From Python to Numpy and in my understanding re-using the existing buffer would avoid such copies.
Question
Would like to understand why there is no out option equivalent in Tensorflow. For instance matmul appear to have no such option to specify the location.Is it because by design Tensorflow will avoid making temporary copies or does it always create temporary copies.
It appears there is no copy indexing or view indexing concepts that numpy has. When an array is extracted from an existing array, is it a shallow copy (view) or a deep copy or it depends?
Please advise where to look at to understand the internal behavior overview similar to From Python to Numpy that gives good insights into its internal architecture and performance considerations.
Tensorflow produces computations graphs, which are highly optimized in terms of the data flow. For example, if some of the stated computations are not needed to produce the final result, TF would not evaluate them. Moreover, TF compiles procedures to its own low-level operations. Hence out parameter of numpy does not make sense in this context.
Thus, TF internally optimizes all steps of the dataflow, and you do not need to provide any instructions. You can optimize the procedure of getting the result as an algorithm, but not how the algorithmworks internally.
To get familiar with the idea what a computational graph is, consider reading this guide
I started by Tensorflow journey when it already came to 2.0.0, So never used graphs and sessions as in version1. But recently met tf.function and autographs which suits me. (but what i know is it is used only for train step)
Now when reading project code, many people use tf.function decorator on many other functions when they wanna build graphs. But i don't exactly get their point. How to know when to use graph and when not?
Can anyone help me?
Solution
The decorator, #tf.function conveniently converts a python function to a static tensorflow graph. TensorFlow operates in eager mode by default since version 2.0.0. Although eager mode could help you in line-by-line execution, this comes with the pitfall of relatively slower TensorFlow-code execution when compared to static-graph. Converting a certain function into a static graph increases execution speed while training your model.
Quoting tf.function documentation:
Functions can be faster than eager code, especially for graphs with many small ops. But for graphs with a few expensive ops (like convolutions), you may not see much speedup.
The static graph is created once and does not get updated if the function is called repeatedly with different values (not passed as the input-arguments). You should avoid using #tf.function in such scenarios or update the function definition (if possible) to include all the necessary variability through the input-arguments. However,
Now, if your function gets all its inputs through the function arguments, then if you apply #tf.function you will not see any problem.
Here is an example.
### When not to use #tf.function ###
# some variable that changes with time
var = timestamp()
#tf.function
def func(*args, **kwargs):
# your code
return var
In the example above, the function func() although depends on var, it does not access the variable var through its arguments. Thus, when #tf.function is applied for the first time, it creates a static-graph for func(). However, when the value of var changes in future, this will not get updated in the static-graph. See this for more clarity. Also, I would highly encourage you to see the references section.
For Debugging
Quoting source
You can use tf.config.experimental_run_functions_eagerly (which temporarily disables running functions as functions) for debugging purposes.
References
Better performance with tf.function
When to utilize tf.function
TensorFlow 2.0: tf.function and AutoGraph
In one deep learning notes (Stanford cs20si), I once saw the following statement regarding eager. I don't quite understand what does the imperative custom layers indicate, and how to understand this code example in the context of imperative custom layers?
Normally, using tensorflow you are not able to access the content of a tensor directly. This means, that you are not able to use if-statements. Instead, you have to construct both possible branches of the branch and then use tf.conditional to include a node which switches between these two, depending on the content of a tensor. This makes it sometimes hard to implement imperative commands in layers.
The example you posted above show, that you are now (with eager execution) able to access the content of tensors, which means, that you can write all the if-statements, for-loops and so on, directly in python and you do not have to construct a huge graph on your own for each possibility. As the code inside the layer is now executed just like a normal imperative programming language, you can call this kind of layer an imperative layer - this is identical to the motivation behind PyTorch.
When I was learning tensorflow, one basic concept of tensorflow was computational graphs, and the graphs was said to be static.
And I found in Pytorch, the graphs was said to be dynamic.
What's the difference of static Computational Graphs in tensorflow and dynamic Computational Graphs in Pytorch?
Both frameworks operate on tensors and view any model as a directed acyclic graph (DAG), but they differ drastically on how you can define them.
TensorFlow follows ‘data as code and code is data’ idiom. In TensorFlow you define graph statically before a model can run. All communication with outer world is performed via tf.Session object and tf.Placeholder which are tensors that will be substituted by external data at runtime.
In PyTorch things are way more imperative and dynamic: you can define, change and execute nodes as you go, no special session interfaces or placeholders. Overall, the framework is more tightly integrated with Python language and feels more native most of the times. When you write in TensorFlow sometimes you feel that your model is behind a brick wall with several tiny holes to communicate over. Anyways, this still sounds like a matter of taste more or less.
However, those approaches differ not only in a software engineering perspective: there are several dynamic neural network architectures that can benefit from the dynamic approach. Recall RNNs: with static graphs, the input sequence length will stay constant. This means that if you develop a sentiment analysis model for English sentences you must fix the sentence length to some maximum value and pad all smaller sequences with zeros. Not too convenient, huh. And you will get more problems in the domain of recursive RNNs and tree-RNNs. Currently Tensorflow has limited support for dynamic inputs via Tensorflow Fold. PyTorch has it by-default.
Reference:
https://medium.com/towards-data-science/pytorch-vs-tensorflow-spotting-the-difference-25c75777377b
https://www.reddit.com/r/MachineLearning/comments/5w3q74/d_so_pytorch_vs_tensorflow_whats_the_verdict_on/
Both TensorFlow and PyTorch allow specifying new computations at any point in time. However, TensorFlow has a "compilation" steps which incurs performance penalty every time you modify the graph. So TensorFlow optimal performance is achieved when you specify the computation once, and then flow new data through the same sequence of computations.
It's similar to interpreters vs. compilers -- the compilation step makes things faster, but also discourages people from modifying the program too often.
To make things concrete, when you modify the graph in TensorFlow (by appending new computations using regular API, or removing some computation using tf.contrib.graph_editor), this line is triggered in session.py. It will serialize the graph, and then the underlying runtime will rerun some optimizations which can take extra time, perhaps 200usec. In contrast, running an op in previously defined graph, or in numpy/PyTorch can be as low as 1 usec.
In tensorflow you first have to define the graph, then you execute it.
Once defined you graph is immutable: you can't add/remove nodes at runtime.
In pytorch, instead, you can change the structure of the graph at runtime: you can thus add/remove nodes at runtime, dynamically changing its structure.
I am using Keras with tensorflow backend and I am curious whether it is possible to skip a layer during backpropagation but have it execute in the forward pass. So here is what I mean
Lambda (lambda x: a(x))
I want to apply a to x in the forward pass but I do not want a to be included in the derivation when the backprop takes place.
I was trying to find a solution bit I could not find anything. Can somebody help me out here?
UPDATE 2
In addition to tf.py_func, there is now an official guide on how to add a custom op.
UPDATE
See this question for an example of writing a custom op with gradient purely in Python without needing to rebuild anything. Note that there are some limitations to the method (see the documentation of tf.py_func).
Not exactly a solution to the problem, but still kind of an answer and too long for comments.
That's not even a Keras issue, but a TensorFlow one. Each op defines its own gradient computation that is used during backpropagation. I you really wanted to something like that, you would need to implement the op into TensorFlow yourself (no easy feat) and define the gradient that you want - because you can't have "no gradient", if anything it would be 1 or 0 (otherwise you can't go on with backpropagation). There is a tf.NoGradient function in TensorFlow which causes an op to propagate zeros, but I don't think it is meant to / can be used out of TensorFlow own internals.
UPDATE
Okay so a bit more of context. TensorFlow graphs are built of ops, which are implemented by kernels; this is basically a 1-to-1 mapping, except that there may be for example a CPU and a GPU kernel for an op, hence the differentiation. The set of ops supported by TensorFlow is usually static, I mean it can change with newer versions, but in principle you cannot add your own ops, because the ops of a graph go into the Protobuf serialized format, so if you made your own ops then you would not be able to share your graph. Ops are then defined at C++ level with the macro REGISTER_OP (see for example here), and kernels with REGISTER_KERNEL_BUILDER (see for example here).
Now, where do gradients come into play? Well, the funny thing is that the gradient of an op is not defined at C++ level; there are ops (and kernels) that implement the gradient of other ops (if you look at the previous files you'll find ops/kernels with the name ending in Grad), but (as far as I'm aware) these are not explicitly "linked" at this level. It seems that the associations between ops and their gradients is defined in Python, usually via tf.RegisterGradient or the aforementioned tf.NoGradient (see for example here, Python modules starting with gen_ are autogenerated with the help of the C++ macros); these registrations inform the backpropagation algorithm about how to compute the gradient of the graph.
So, how to actually work this out? Well, you need to create at least one op in C++ with the corresponding kernel/s implementing the computation that you want for your forward pass. Then, if the gradient computation that you want to use can be expressed with existing TensorFlow ops (which is most likely), you would just need to call tf.RegisterGradient in Python and do the computation there in "standard" TensorFlow. This is quite complicated, but the good news is it's possible, and there's even an example for it (although I think they kinda forgot the gradient registration part in that one)! As you will see, the process involves compiling the new op code into a library (btw I'm not sure if any of this may work on Windows) that is then loaded from Python (obviously this involves going through the painful process of manual compilation of TensorFlow with Bazel). A possibly more realistic example can be found in TensorFlow Fold, an extension of TensorFlow for structured data that register (as of one) one custom operation here through a macro defined here that calls REGISTER_OP, and then in Python it loads the library and register its gradient here through their own registration function defined here that simply calls tf.NotDifferentiable (another name for tf.NoGradient)
tldr: It is rather hard, but it can be done and there are even a couple of examples out there.
As mentioned in #jdehesa's comments. You can implement your function with an "alternative gradient". Forgive me if my math is not correct, but I think a derivative returning "1" would be the correct way to have no effect on the backpropagation while still passing the learning through. For how to construct it, see here. The example I cited goes further and allows you to construct an activation function from a python function. So in place of the spiky function, substitute your function a, and in place of his derivative d_spiky replace it with
def constant(x):
return 1
So on the forward pass, a is applied in the layer and the the backwards pass 1 is applied which should simply pass the weight adjustments through.
You can then just create an Activation layer in Keras using this function.