matlab - Training neural network tips -

-

The object is proposed for validation that I want to use a neural network in MATLAB. I have 30 objects and 20 images for each object, so I have 600 input data and 20 different classes I input matrix is ​​100x600 and the target is 1x600. The input matrix column is a histogram of 100 compartments like the keypoint 'hu': (m, n) = hist (hue_val, 100) i got m .
If I have chosen the MLP network then how many layers and neurons are required for those layers, which are the appropriate functions for each layer?

And for the last question, do I want negative samples?

  • Number of layers - usually a hidden layer is sufficient (for so long Using your non-linear activation function) One can guess an arbitrary number of layer layers.
  • Function Transfer - I have not been used for this word, but I think your activation function (which you do on the net input before passing it on the next layer). I answered a slight type of question, but it has been said that in most cases a standard option like hyperbolic tangent or logistic work.
  • Number of neurons in your hidden layer - Crudrichzo Link provided it very well in the field. What I can really add is that with the input size, maybe at the training time I would be very much of this quantity. Negative specimens - If you only need to classify one of the 30 items, then negative sample is not required. But, if a test input of 30 objects can be any , then definitely use many negative examples so that the network does not believe everything is an object.

    Some Common Tips:

    Remember to consider leaving one-out and similar forms as a way of dealing with over-fighting. It also limits the unit calculation of your hidden layer, but at the expense of the richness of representation.

    The other parameters you specify are also very important for any successful ANN application. This includes learning rates, error functions, annealing schedule, speed and weight loss. Setting all of these above is more than an art than science (one of the best arguments against using ANN vs. support vector machines), but this is a boon for me in this area.

Comments

Post a Comment

Popular posts from this blog

mysql - BLOB/TEXT column 'value' used in key specification without a key length -

-
I have developed an extension which works up to 1.6 on Magren (I'm trying Enterprise Edition, And I think the community is the same problem, because it is the same code). In my install script, I see the $ installer-> gt; CreateEntityTables ($ this- & gt; getTable ('alphanum / info')); The installation is done until it is not in the _text unit table. It crashed there! It turns out that when I log in to SQL and run it via PHPmyadmin, then this error is: Blob / Text column 'value' is used without the key 'key' . I saw the code there, and this is what is trying to create an index on the value column: -> addIndex ($ this- & gt; getIdxName ($ eavTableName, array ( 'attribute_id array (' attribute_id ',' value ')) - & gt; addIndex ($ this- & gt; getIdxName ($ eavTableName, array (' entity_type_id ' , 'Value')), array ('entity_type_id', 'value')) If there is no if statement is n...
Read more

c# - Using Vici cool Storage with monodroid -

-
Oh, for the moment I am not sure how it is officially supported, but for the success of the people Using Monodroid Vic Cool Storage I am able to leave and compile assemblies in my project, but while trying to use some classes, the errors of time have to be gathered. Especially when trying to connect like the example for Mono Touch string dbName = path. Cobain (environment. Gatefolder path (environment specialfolder personal), "mydb.db3"); // The following line will be called coolestation, where is the database, // to make it if it does not exist, and call a delegate that creates the necessary tables (only when the database file was newly created //) CSConfig.SetDB (DbName, true, () => {CSDatabase.ExecuteNonQuery ("Subscriber Integer Primary Key Attribute, Name Text (50) No, Number, Date First Text (30)");); There is no difference when I try to use the methods of CSConfig and when I cscfif If I try to pass 3 aggregates to STDB (3) then I get the invalid ...
Read more

python - referencing a variable in another function? -

-
I am an experienced TCL developer and write my own processes to help myself. I call it a proc i callVar, print out the value of the variable in a certain format, so I know that it is variable and what is the value "set foo 1; putVar foo" result 'foo = "1" I want to do the same thing in Python, but I can not find the answer :( In TCL I use the up command because when I pass the name of the variable, then I can see the value at the top. . I understand that an advanced light in the dragon There is no mechanism of r, but with its introspection capability, I would think that this is something trivial, but I am not getting the answer anywhere. It seems that this should work: and expanded: def foo (): y = "Hello World" printwor ("y") foo () y = "Hello World " You can, but you should not do so if your code is correct If is needed , then there is something wrong with your design, and you should probably use the dictionar...
Read more