"Fossies" - the Fresh Open Source Software Archive

Member "pytorch-1.8.2/caffe2/observers/README.md" (23 Jul 2021, 1459 Bytes) of package /linux/misc/pytorch-1.8.2.tar.gz:

As a special service "Fossies" has tried to format the requested source page into HTML format (assuming markdown format). Alternatively you can here view or download the uninterpreted source code file. A member file download can also be achieved by clicking within a package contents listing on the according byte size field.



Observers are a small framework that allow users to attach code to the execution of SimpleNets and Operators.

An example of an Observer is the TimeObserver, used as follows:


unique_ptr<TimeObserver<NetBase>> net_ob =
auto* ob = net->AttachObserver(std::move(net_ob));
LOG(INFO) << "av time children: " << ob->average_time_children();
LOG(INFO) << "av time: " << ob->average_time();


ob = model.net.GetObserver("TimeObserver")

print("av time children:", ob.average_time_children())
print("av time:", ob.average_time())

Histogram Observer

Creates a histogram for the values of weights and activations

                      "histogram.txt", # filename
                      2014, # number of bins in histogram
                      32 # Dumping frequency

This will generate a histogram for the activations and store it in histogram.txt

Implementing An Observer

To implement an observer you must inherit from ObserverBase and implement the Start and Stop functions.

Observers are instantiated with a subject of a generic type, such as a Net or Operator. The observer framework is built to be generic enough to "observe" various other types, however.