These are another new questions after solving the previous question,"How to classify voice data?". Thank anyone for answering any question.

Now I have three questions (shown below with green background) when I try to build transfer learning model.

1. I want to add Normailize in my net, but I can't find Normalize[Exp[..],Total] which is hinted by the document. How to add Normalize[Exp[..],Total] ? I only find BatchNormalizationLayer here.

Below questions I have referred to two official examples. But I can't slove them.

2. Precompute the features using the feature extractor and train only the classifier net. I can't find an example of this method to build my transfer learning model.

3. How to solve the problem from the NetTrain?

Transfer Learning

Obtain the “Audio Cats and Dogs” dataset and divide it into training and test sets of approximately equal cumulative duration.

In[]:=

ResourceObject["Audio Cats and Dogs"];​​{dogDuration, catDuration} = Values@Normal[​​ GroupBy[ResourceData["Audio Cats and Dogs"], "Label"][All, Mean@*Duration, "Audio"]];​​​​SeedRandom[42];​​classLength=93;​​{catTrain,catTest}=TakeDrop[RandomSample[ResourceData["Audio Cats and Dogs"][Select[#Label==="cat"&]]],Round[classLength*dogDuration/catDuration]];​​{dogTrain,dogTest}=TakeDrop[RandomSample[ResourceData["Audio Cats and Dogs"][Select[#Label==="dog"&]]],classLength];​​​​balancedTrain = RandomSample[Join[catTrain, dogTrain]];​​​​SeedRandom[42];​​catTestex=RandomSample[catTest,10];​​​​balancedTest = RandomSample[Join[dogTest, catTestex]];

Now for the transfer learning.... Create the AudioIdentify feature extractor net which will be used as the starting point for training.

Hint:Extractthecore

AudioIdentify

net,​​chopofftheclassificationlayersandreconstructthefullnetusing

NetMapOperator

tocomputefeaturesoneachchunkand

AggregationLayer

toaveragethesesequencesoverthetimedimension.Theoutputshouldbeasingleflatvectoroflength1664.

In[]:=

net=NetModel["Wolfram AudioIdentify V1 Trained on AudioSet Data","Size""Large"];​​mainNet=NetExtract[net,{1,"Net"}];​​featureExtractor=​​ NetChain[NetMapOperator[NetDrop[mainNet,-3]],​​ AggregationLayer[Mean, 1], FlattenLayer[],"Input"net[["Input"]]]

Out[]=

NetChain



Input port: expression Output port: vector (size: 1664)



Create the net that will be in charge of classifying cats and dogs.

Try to implement the following structure:• Start with a form of regularization (such as Dropout) applied to the incoming features (to avoid overfitting the small dataset and to make the net’s prediction more stable by averaging the predictions of an ensemble of smaller nets).• Project the learned features onto a smaller space (try a magnitude similar to the number of classes) to perform classification.​

• Apply

Normalize

to the final output vector so that it can be interpreted as a probability distribution (use a layer that implements the

equivalent of

Normalize[Exp[..], Total]

).

​

• Decode the probability distribution into class labels using a

NetDecoder

.

In[]:=

leakyRelu = ​​ ParametricRampLayer[{}, "Slope" -> 0.1, LearningRateMultipliers -> 0,​​ "Input" -> 2]

Out[]=

ParametricRampLayer



Input : vector (size: 2) Output : vector (size: 2)



In[]:=

classifier = NetChain[{DropoutLayer[0.2],leakyRelu,BatchNormalizationLayer["Input" -> 2],SoftmaxLayer[]}, "Output" -> NetDecoder[{"Class", {"dog", "cat"}}]]

Out[]=

NetChain



uniniti aliz ed

Input vector (size: 2) 1 DropoutLayer vector (size: 2) 2 ParametricRampLayer vector (size: 2) 3 BatchNormalizationLayer vector (size: 2) 4 SoftmaxLayer vector (size: 2) Output class



Train the classification layers for the new task.

Hint: There are a couple different ways to accomplish this:​

1. Precompute the features using the feature extractor and train only the classifier net.

​

( An advantage of this approach is that you can save the features in

WXF

format.)​​

2. Train the full joint feature extractor + classifier net with

LearningRateMultipliers

specified to avoid relearning weights in layers except for the final

classification layers.

In[]:=

trainedNet = NetTrain[classifier, balancedTrain,LossFunction -> "Loss2"]

NetTrain

:Training data specification should include values for the ports "Input" and "Output"​, but only contains values for the ports "Audio" and "Label". You may need to explicitly specify the loss port(s) of the net using LossFunction.

Out[]=

$Failed