Différences

Ci-dessous, les différences entre deux révisions de la page.

--- keras_input_explanation [2020/10/03 11:08] – [Shapes of the tensors] serge
+++ keras_input_explanation [2020/12/27 15:14] (Version actuelle) – serge
@@ Ligne 1: / Ligne 1: @@
 ====== Keras input explanation: input_shape, units, batch_size, dim, etc ======
+<WRAP center round box 60% centeralign>
+**{{tagpage>ia|Intelligence Artificielle}}**     **[[http://translate.google.com/translate?hl=&sl=auto&tl=en&u=https%3A%2F%2Fressources.labomedia.org%2Fkeras_input_explanation|English Version]]**
+</WRAP>
+<WRAP center round box 60% centeralign>
+**[[les_pages_intelligence_artificielle_en_details|Les Pages Intelligence Artificielle en détails]]**
+</WRAP>
 Mise en forme d'un post de **stackoverflow.com** non traduit en français, car très technique ce qui ne se comprend qu'en anglais.
 **[[https://stackoverflow.com/questions/44747343/keras-input-explanation-input-shape-units-batch-size-dim-etc|Keras input explanation: input_shape, units, batch_size, dim, etc]]**
+=====Ressources=====
+  * [[https://keras.io/guides/working_with_rnns/|Working with RNNs]]
+  * [[https://keras.io/api/layers/|Keras layers API]]
+  * [[https://keras.io/examples/|Code examples]]
 =====Question=====
@@ Ligne 13: / Ligne 25: @@
 {{ :media_10:input_explanation.jpg?600 |}}
-=====Réponse 1=====
+=====Définitions=====
-===Units===
+====Units====
-The amount of "neurons", or "cells", or whatever the layer has inside it.
+The amount of "neurons", or "cells", or whatever the layer has inside it. \\
 It's a property of each layer, and yes, it's related to the output shape (as we will see later). In your picture, except for the input layer, which is conceptually different from other layers, you have:
   * Hidden layer 1: 4 units (4 neurons)
   * Hidden layer 2: 4 units
   * Last layer: 1 unit
-  *
-===Shapes===
+====Shapes====
-Shapes are consequences of the model's configuration. Shapes are tuples representing how many elements an array or tensor has in each dimension.
+Shapes are consequences of the model's configuration. Shapes are tuples representing how many elements an array or tensor has in each dimension.\\
 Ex: a shape (30,4,10) means an array or tensor with 3 dimensions, containing 30 elements in the first dimension, 4 in the second and 10 in the third, totaling 30*4*10 = 1200 elements or numbers.
-===The input shape===
+====The input shape====
-What flows between layers are tensors. Tensors can be seen as matrices, with shapes.
+What flows between layers are tensors. Tensors can be seen as matrices, with shapes. \\
-In Keras, the input layer itself is not a layer, but a tensor. It's the starting tensor you send to the first hidden layer. This tensor must have the same shape as your training data.
+In Keras, the input layer itself is not a layer, but a tensor. It's the starting tensor you send to the first hidden layer. **This tensor must have the same shape as your training data**. \\
-Example: if you have 30 images of 50x50 pixels in RGB (3 channels), the shape of your input data is (30,50,50,3). Then your input layer tensor, must have this shape (see details in the "shapes in keras" section).
+Example: if you have 30 images of 50x50 pixels in RGB (3 channels), the shape of your input data is (30,50,50,3). Then your input layer tensor, must have this shape (see [[keras_input_explanation#shapes_in_keras| Shapes in Keras]]). \\
 Each type of layer requires the input with a certain number of dimensions:
   * Dense layers require inputs as (batch_size, input_size)
@@ Ligne 37: / Ligne 49: @@
     *if using channels_first: (batch_size, channels, imageside1, imageside2)
   * 1D convolutions and recurrent layers use (batch_size, sequence_length, features)
-    *Details on how to prepare data for recurrent layers
+    *Details on [[https://stackoverflow.com/questions/38714959/understanding-keras-lstms/50235563#50235563|how to prepare data for recurrent layers ]]
-Now, the input shape is the only one you must define, because your model cannot know it. Only you know that, based on your training data.
+Now, the input shape is the only one you must define, because your model cannot know it. Only you know that, based on your training data. \\
 All the other shapes are calculated automatically based on the units and particularities of each layer.
-===Relation between shapes and units - The output shape===
+====Relation between shapes and units - The output shape====
-Given the input shape, all other shapes are results of layers calculations.
+Given the input shape, all other shapes are results of layers calculations. \\
-The "units" of each layer will define the output shape (the shape of the tensor that is produced by the layer and that will be the input of the next layer).
+The "units" of each layer will define the output shape (the shape of the tensor that is produced by the layer and that will be the input of the next layer). \\
-Each type of layer works in a particular way. Dense layers have output shape based on "units", convolutional layers have output shape based on "filters". But it's always based on some layer property. (See the documentation for what each layer outputs)
+Each type of layer works in a particular way. Dense layers have output shape based on "units", convolutional layers have output shape based on "filters". But it's always based on some layer property. (See the documentation for what each layer outputs) \\
-Let's show what happens with "Dense" layers, which is the type shown in your graph.
+Let's show what happens with "Dense" layers, which is the type shown in your graph. \\
 A dense layer has an output shape of (batch_size,units). So, yes, units, the property of the layer, also defines the output shape.
   * Hidden layer 1: 4 units, output shape: (batch_size,4).
@@ Ligne 51: / Ligne 64: @@
   * Last layer: 1 unit, output shape: (batch_size,1).
-===Weights===
+====Weights====
-Weights will be entirely automatically calculated based on the input and the output shapes. Again, each type of layer works in a certain way. But the weights will be a matrix capable of transforming the input shape into the output shape by some mathematical operation.
+Weights will be entirely automatically calculated based on the input and the output shapes. Again, each type of layer works in a certain way. But the weights will be a matrix capable of transforming the input shape into the output shape by some mathematical operation. \\
-In a dense layer, weights multiply all inputs. It's a matrix with one column per input and one row per unit, but this is often not important for basic works.
+In a dense layer, weights multiply all inputs. It's a matrix with one column per input and one row per unit, but this is often not important for basic works. \\
 In the image, if each arrow had a multiplication number on it, all numbers together would form the weight matrix.
-===Shapes in Keras===
+====Shapes in Keras====
-Earlier, I gave an example of 30 images, 50x50 pixels and 3 channels, having an input shape of (30,50,50,3).
+Earlier, I gave an example of 30 images, 50x50 pixels and 3 channels, having an input shape of (30,50,50,3). \\
 Since the input shape is the only one you need to define, Keras will demand it in the first layer.
 But in this definition, Keras ignores the first dimension, which is the batch size. Your model should be able to deal with any batch size, so you define only the other dimensions:
-input_shape = (50,50,3)
-    #regardless of how many images I have, each image has this shape
+input_shape = (50,50,3) # regardless of how many images I have, each image has this shape
-Optionally, or when it's required by certain kinds of models, you can pass the shape containing the batch size via batch_input_shape=(30,50,50,3) or batch_shape=(30,50,50,3). This limits your training possibilities to this unique batch size, so it should be used only when really required.
-Either way you choose, tensors in the model will have the batch dimension.
+Optionally, or when it's required by certain kinds of models, you can pass the shape containing the batch size via batch_input_shape=(30,50,50,3) or batch_shape=(30,50,50,3). This limits your training possibilities to this unique batch size, so it should be used only when really required.\\
-So, even if you used input_shape=(50,50,3), when keras sends you messages, or when you print the model summary, it will show (None,50,50,3).
+Either way you choose, tensors in the model will have the batch dimension.\\
-The first dimension is the batch size, it's None because it can vary depending on how many examples you give for training. (If you defined the batch size explicitly, then the number you defined will appear instead of None)
+So, even if you used input_shape=(50,50,3), when keras sends you messages, or when you print the model summary, it will show (None,50,50,3).\\
-Also, in advanced works, when you actually operate directly on the tensors (inside Lambda layers or in the loss function, for instance), the batch size dimension will be there.
+The first dimension is the batch size, it's None because it can vary depending on how many examples you give for training. (If you defined the batch size explicitly, then the number you defined will appear instead of None)\\
+Also, in advanced works, when you actually operate directly on the tensors (inside Lambda layers or in the loss function, for instance), the batch size dimension will be there. \\
   * So, when defining the input shape, you ignore the batch size: input_shape=(50,50,3)
   * When doing operations directly on tensors, the shape will be again (30,50,50,3)
   * When keras sends you a message, the shape will be (None,50,50,3) or (30,50,50,3), depending on what type of message it sends you.
-===Dim===
+====Dim====
-And in the end, what is dim?
+And in the end, what is dim? \\
-If your input shape has only one dimension, you don't need to give it as a tuple, you give input_dim as a scalar number.
+If your input shape has only one dimension, you don't need to give it as a tuple, you give input_dim as a scalar number. \\
 So, in your model, where your input layer has 3 elements, you can use any of these two:
   * input_shape=(3,) -- The comma is necessary when you have only one dimension
@@ Ligne 79: / Ligne 93: @@
 But when dealing directly with the tensors, often dim will refer to how many dimensions a tensor has. For instance a tensor with shape (25,10909) has 2 dimensions.
-====Defining your image in Keras====
+=====Defining your image in Keras=====
 Keras has two ways of doing it, Sequential models, or the functional API Model. I don't like using the sequential model, later you will have to forget it anyway because you will want models with branches.
 PS: here I ignored other aspects, such as activation functions.
@@ Ligne 89: / Ligne 103: @@
 model = Sequential()
-#start from the first hidden layer, since the input is not actually a layer
+# Start from the first hidden layer, since the input is not actually a layer
-#but inform the shape of the input, with 3 elements.
+# but inform the shape of the input, with 3 elements.
-model.add(Dense(units=4,input_shape=(3,))) #hidden layer 1 with input
+model.add(Dense(units=4,input_shape=(3,))) # hidden layer 1 with input
-#further layers:
+# Further layers:
-model.add(Dense(units=4)) #hidden layer 2
+model.add(Dense(units=4)) # hidden layer 2
-model.add(Dense(units=1)) #output layer
+model.add(Dense(units=1)) # output layer
-With the functional API Model:
+</code>
+**With the functional API Model:**
+<code python>
 from keras.models import Model
 from keras.layers import *
-#Start defining the input tensor:
+# Start defining the input tensor:
 inpTensor = Input((3,))
-#create the layers and pass them the input tensor to get the output tensor:
+# Create the layers and pass them the input tensor to get the output tensor:
 hidden1Out = Dense(units=4)(inpTensor)
 hidden2Out = Dense(units=4)(hidden1Out)
 finalOut = Dense(units=1)(hidden2Out)
-#define the model's start and end points
+# Define the model's start and end points
-model = Model(inpTensor,finalOut)
+model = Model(inpTensor, finalOut)
 </code>
 ====Shapes of the tensors====
-Remember you ignore batch sizes when defining layers:
+Remember you ignore batch sizes when defining layers, so set it to None:
   * inpTensor: (None,3)
   * hidden1Out: (None,4)
@@ Ligne 120: / Ligne 138: @@
 ====Comments 1====
-  * One question about the input_shape= parameter remains: to which dimension the first value of the argument refers? I see things like input_shape=(728, ), so in my mind the first argument refers to columns (fixed) and second to rows (free to vary). But how does this sit with Python's row-major order of arrays?
+To which dimension the first value of the argument refers? I see things like input_shape=(728, ), so in my mind the first argument refers to columns (fixed) and second to rows (free to vary). But how does this sit with Python's row-major order of arrays?
   * That comma does not create a second dimension. It's just python notation for creating a tuple that contains only one element. input_shape(728,) is the same as batch_input=(batch_size,728). This means that each sample has 728 values.
-  * @DanielMöller: could you please elaborate a little bit what the difference between "input elements" and "dimensions" are? I would think that the graph above had a three-dimensional input layer, thus making dim=3, so I'm wondering what I'm missing here, because I see you write that the input is 1-dimensional...
-  * A vector has one dimension, but many elements. It has shape (n,) ---- A matrix has two dimensions, dimension 0 has m elements, dimension 1 has n elements, totaling m x n elements, shape (m,n). If you imagine a "cube" divided in little cubes, each little cube with data, this would be 3D, with m x n x o elements, shape (m,n,o).
+===Q===
-  * But when dealing directly with the tensors, often dim will refer to how many dimensions a tensor has. So for 1-dimensional 3-length input, input_dim=3, but for 2-dimensional input of shape (25,10909), input_dim=2?
+Could you please elaborate a little bit what the difference between "input elements" and "dimensions" are? I would think that the graph above had a three-dimensional input layer, thus making dim=3, so I'm wondering what I'm missing here, because I see you write that the input is 1-dimensional...
+  * A vector has one dimension, but many elements. It has shape (n,) ---- A matrix has two dimensions, dimension 0 has m elements, dimension 1 has n elements, totaling m x n elements, shape (m,n). If you imagine a "cube" divided in little cubes, each little cube with data, this would be 3D, with m x n x o elements, shape (m,n,o).
+===Q===
+But when dealing directly with the tensors, often dim will refer to how many dimensions a tensor has. So for 1-dimensional 3-length input, input_dim=3, but for 2-dimensional input of shape (25,10909), input_dim=2?
   * input_shape=(25,10909)
   * For those who have inputs and output tensors of different dimensions, see Daniel's other answer here: stackoverflow.com/questions/46404581/… tl;dr model.add(Flatten())
-  * question @DanielMöller - you said if you have 30 images of 50x50 pixels in RGB (3 channels), the shape of your input data is (30,50,50,3) how do you get the intuition behind this? Shouldn't it be(30*3*50*50) 50*50 for each image then *3 for RGB then *30 for examples. OR is it that the dimensions don't matter and using any one convention will do?
-  * @Prince, the order matters. You can configure Keras to use data_format = 'channels_first' or data_format='channels_last'. I recommend using always channels last (Keras' default). It's more compatible with all other layers.
+===Q===
-  * "or (batch_size, optional,...,optional, input_size)" - in what circumstances would you pass the optional data items here? What are they?
+You said if you have 30 images of 50x50 pixels in RGB (3 channels), the shape of your input data is (30,50,50,3) how do you get the intuition behind this? Shouldn't it be(30*3*50*50) 50*50 for each image then *3 for RGB then *30 for examples. OR is it that the dimensions don't matter and using any one convention will do?
-  * @DuncanJones, they're whatever you want, if your data has this format. The Dense layer will only work on the last dimension, leaving the other dimensions untouched.
+  * the order matters. You can configure Keras to use data_format = 'channels_first' or data_format='channels_last'. I recommend using always channels last (Keras' default). It's more compatible with all other layers.
+===Q===
+"or (batch_size, optional,...,optional, input_size)" - in what circumstances would you pass the optional data items here? What are they?
+  * They're whatever you want, if your data has this format. The Dense layer will only work on the last dimension, leaving the other dimensions untouched.
   * Examples, video frames that should be treated equally. Image pixels when you want to process only the color channels for each pixel instead of mixing pixels, etc.
-  * @DanielMöller Thanks a lot for the great answer. I have a related question about the input to keras: Please kindly let me know your thoughts on this :)
-  * not sure someone will see till this point, but, anyway, if you are using cv2.resize for resizing then you should keep in mind that the output has height , width and Input(shape = (width, height, channels))
+=====Input Dimension Clarified=====
+It (the word dimension alone) can refer to:
+  - **The dimension of Input Data (or stream)** such as # N of sensor axes to beam the time series signal, or RGB color channel (3): suggested word=> "InputStream Dimension"
+  - **The total number /length of Input Features** (or Input layer) (28 x 28 = 784 for the MINST color image) or 3000 in the FFT transformed Spectrum Values, or "Input Layer / Input Feature Dimension"
+  - **The dimensionality** (# of dimension) of the input (typically 3D as expected in Keras LSTM) or (#RowofSamples, #of Senors, #of Values..) 3 is the answer.
+  - "N Dimensionality of Input" **The SPECIFIC Input Shape** (eg. (30,50,50,3) in this unwrapped input image data, or (30, 250, 3) if unwrapped Keras:
+Keras has its input_dim refers to the Dimension of Input Layer / Number of Input Feature
+<code python>
+model = Sequential()
+model.add(Dense(32, input_dim=784))  #or 3 in the current posted example above
+model.add(Activation('relu'))
+</code>
+In Keras LSTM, it refers to the total Time Steps
+The term has been very confusing, is correct and we live in a very confusing world!!
+I find one of the challenge in Machine Learning is to deal with different languages or dialects and terminologies (like if you have 5-8 highly different versions of English, then you need to very high proficiency to converse with different speakers). Probably this is the same in programming languages too.
-{{tag>sb ia}}
+{{tag>sb ia keras}}