with(Jupyter);

$\left[\mathrm{CreateNotebook}\,\mathrm{ExtractCodeSources}\,\mathrm{GenerateKernelConfiguration}\,\mathrm{SetOutputRendererByType}\right]$

CreateNotebook( "Connectivity.ipynb", "updates,Maple2022,Connectivity", source=help, base=homedir );

$20615$

restart:

with(DeepLearning):

M := RandomTensor( Gamma(2.5,3.3), [5,4,3], datatype=float[8], seed=2022 );

$M≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [5,\; 4,\; 3]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

M2 := M[1..2,..,1..2];

$\mathrm{M2}≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [2,\; 4,\; 2]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

M[2,3,2];

$10.7615426222778$

convert( M, Array );

A GradientTape is a execution context in which certain marked Tensors are watched for the purposes of computing their gradients.

u := Constant([[3., 5.]], datatype=float[8]);

$u≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [1,\; 2]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

The following creates a tape object to track variables of interest.

tape := GradientTape();

$\mathrm{tape}≔\left[\begin{array}{c}\mathrm{DeepLearning\; GradientTape}\\ \mathrm{<tensorflow.python.eager.backprop.GradientTape\; object\; at\; 0x0000022F309EC730>}\end{array}\right]$

The Enter and Exit commands activate and deactivate this context. Only operations which occur after Enter has been invoked are tracked.

Enter( tape );

$\left[\begin{array}{c}\mathrm{DeepLearning\; GradientTape}\\ \mathrm{<tensorflow.python.eager.backprop.GradientTape\; object\; at\; 0x0000022F309EC730>}\end{array}\right]$

tape:-Watch(u);

v := u^2;

$v≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [1,\; 2]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

Exit( tape );

grad := tape:-Gradient( v, u );

$\mathrm{grad}≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [1,\; 2]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

convert( grad, Matrix );

$\left[\begin{array}{cc}6.& 10.\end{array}\right]$

As DeepLearning is built on the Google TensorFlow library using Maple's connectivity to Python, there is a close connection between many DeepLearning objects and TensorFlow objects. The convert command now permits easy conversion between DeepLearning objects and objects of type python from TensorFlow. This will simplify the adaptation of TensorFlow models to DeepLearning and facilitate the interaction of DeepLearning objects with TensorFlow features which have not been exposed in DeepLearning.

The command convert(..., python) converts DeepLearning objects to their Python TensorFlow equivalents.

t1 := Constant( <<1,2,3>|<4,5,6>|<7,8,9>>, datatype=float[8] );

$\mathrm{t1}≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [3,\; 3]}\\ \mathrm{Data\; Type:\; float[8]}\end{array}\right]$

convert( t1, python );

Python:-ImportModule("tensorflow as tf");

The command convert(..., DeepLearning) converts DeepLearning objects to their Python TensorFlow equivalents.

pt2 := Python:-EvalString( "tf.random.uniform(shape=[3,4], maxval=20, dtype=tf.int32, seed=1701)" );

t2 := convert( pt2, DeepLearning );

$\mathrm{t2}≔\left[\begin{array}{c}\mathrm{DeepLearning\; Tensor}\\ \mathrm{Shape:\; [3,\; 4]}\\ \mathrm{Data\; Type:\; integer[4]}\end{array}\right]$

pt3 := Python:-EvalFunction( "tf.keras.layers.GaussianNoise", 5.5 );

t3 := convert( pt3, DeepLearning );

$\mathrm{t3}≔\left[\begin{array}{c}\mathrm{DeepLearning\; Layer}\\ \mathrm{<tensorflow.python.keras.layers.noise.GaussianNoise\; object\; at\; 0x0000022F309D56A0>}\end{array}\right]$

with(ImageTools):

qrCode := GenerateBarcode( "This is a test" );

ImageTools:-Embed( Scale( qrCode, 9, method=nearest ) );

The convert command to objects of type Array, Matrix, Vector, string and ByteArray has been supplemented with an additional option, sourceformat. This reads encoded data from a string or ByteArray in a specified to the named type with no need for external files.

Here we can read directly from a comma-delimited string to a Matrix:

convert( "1,2\n3,4", Matrix, sourceformat="CSV" );

$\left[\begin{array}{cc}1& 2\\ 3& 4\end{array}\right]$

The format specified must be one accepted by the Import command. When the output type is not string or ByteArray, sourceformat may be abbreviated simply as format.

Additionally, the convert command to objects of type string and ByteArray has been supplemented with an additional option, targetformat. This writes the encoded data to the output string or ByteArray in the named format, which must be one accepted by the Export command.

This has the same effect as exporting the content to a file in the chosen format with Export and then reading it to a string or ByteArray with FileTools:-Text:-ReadFile or FileTools:-Binary:-ReadFile respectively, but is achieved without any user-visible need for file input or output.

When the input type is not string or ByteArray, targetformat may be abbreviated simply as format.

In these examples we convert expressions to strings encoded in JSON and LaTeX respectively.

T := table([ "firstname" = "G", "lastname" = "Raymond", "DOB" = "1960-02-28" ]);

$T≔table\left(\left[''lastname''=''Raymond''\&comma;''firstname''=''G''\&comma;''DOB''=''1960-02-28''\right]\right)$

convert( T, string, format="JSON" );

$''\{\; "DOB":\; "1960-02-28",\; "firstname":\; "G",\; "lastname":\; "Raymond"\; \}''$

convert( { solve(a*x^2 + b*x + c = 0, x) }, string, format="LaTeX" );

$''\backslash left\backslash \{\backslash frac\{-b\; +\backslash sqrt\{-4\; a\; c\; +b^\{2\}\}\}\{2\; a\},\; -\backslash frac\{b\; +\backslash sqrt\{-4\; a\; c\; +b^\{2\}\}\}\{2\; a\}\backslash right\backslash \}''$

In the following example we convert a 3-D plot to a ByteArray encoded in the STL format

knot := algcurves:-plot_knot((-x^7 + y^3)*(-2*x^5 + y^2), x, y, epsilon = 0.8, radius = 0.1, tubepoints = 9);

convert( knot, ByteArray, targetformat="STL" );

The new MultipartFormPost command in the URL package provides another protocol for uploading data to a web service.  This command mimics the behavior of a submit button on a form-based webpage; usually one asking for a file upload and other input data.  The new command can package all of the entries together, send them to the website in the correct format (either for storage, or analysis), and get the post-submit page back for further processing if needed.