Update course information for Summer Semester 2024 (#9)

* Update general information

* Update lecture and homework headers

* Update homework due dates

* Add course timeline

README.md

@@ -27,7 +27,8 @@ Follow the [installation instructions](https://adrhill.github.io/julia-ml-course
 on the course website.
 
 ## Contents
-The course is taught in five weekly sessions of three hours.
+### Lectures
+The **first half of the course** is taught in five weekly sessions of three hours.
 In each session, two lectures are taught:
 
 | Week | Lecture | Content                                           |
@@ -80,9 +81,20 @@ The lectures and the homework cover the following packages:
 | ProfileView.jl       |       9 | Profiler                                               |
 | Cthulhu.jl           |       9 | Type inference debugger                                |
 
+### Project
+In the **second half of the course**, after passing the homework,
+students work in groups on a small programming project of their choice, 
+learning best practices for package development in Julia, such as:
+* how to structure and develop a package
+* how to write package tests
+* how to write and host package documentation
+
+During code review sessions, students give each other feedback on their projects 
+before presenting their work in end-of-semester presentations.
+
 [site-url]: https://adrhill.github.io/julia-ml-course/
-[ml-group-url]: https://www.tu.berlin/ml
-[isis-url]: https://isis.tu-berlin.de/course/view.php?id=35533
+[ml-group-url]: https://web.ml.tu-berlin.de
+[isis-url]: https://isis.tu-berlin.de/course/view.php?id=37588
 
 [goto-badge]: https://img.shields.io/badge/-Go%20to%20course%20website-informational
 [isis-badge]: https://img.shields.io/badge/TU%20Berlin-ISIS%20page-red

homework/H1_Basics.jl

@@ -37,13 +37,13 @@ html"""
 			Homework 1: Julia Programming Basics
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24<br>
+			TU Berlin, Summer Semester 2024<br>
 		</p>
 	</div>
 """
 
 # ╔═╡ bdcb27c5-0603-49ac-b831-d78c558b31f0
-md"Due date: **Monday, November 27th 2023 at 23:59**"
+md"Due date: **Monday, April 29th 2024 at 23:59**"
 
 # ╔═╡ 6be73c03-925c-4afa-bd66-aca90e6b49fe
 md"This notebook gives you live feedback!

homework/H2_Linear_Algebra.jl

@@ -49,13 +49,13 @@ html"""
 			Homework 2: Indexing & Linear Algebra
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24<br>
+			TU Berlin, Summer Semester 2024<br>
 		</p>
 	</div>
 """
 
 # ╔═╡ bdcb27c5-0603-49ac-b831-d78c558b31f0
-md"Due date: **Monday, December 4th 2023 at 23:59**"
+md"Due date: **Monday, May 6th 2024 at 23:59**"
 
 # ╔═╡ ddd6e83e-5a0d-4ff0-afe4-dedfc860994c
 md"### Student information"
@@ -233,7 +233,7 @@ task(
 
 ```
 If the image contains an uneven number of columns, ignore the center column.
-For example, applying `my_crop` to an input matrix 
+For example, applying `my_crop` to an input matrix
 ```julia
 3×5 Matrix{Int64}:
  1  4  7  10  13
@@ -305,14 +305,14 @@ my_crop(img)
 
 # ╔═╡ 3e933ccb-a6f3-4c2c-99ae-7ff03ed3a786
 md"## Exercise 2: Singular value decomposition
-The [Singular value decomposition](https://en.wikipedia.org/wiki/Singular_value_decomposition) (SVD) factorizes matrices $M$ into 
+The [Singular value decomposition](https://en.wikipedia.org/wiki/Singular_value_decomposition) (SVD) factorizes matrices $M$ into
 
 $M = U \Sigma V' \quad .$
 
-LinearAlgebra.jl implements SVD in the function `svd`. For this implementation of SVD:  
+LinearAlgebra.jl implements SVD in the function `svd`. For this implementation of SVD:
 *  $\Sigma$ is a vector of  sorted, non-negative real numbers, the [singular values](https://en.wikipedia.org/wiki/Singular_value)
 *  $U$ is a unitary matrix
-*  $V$ is a unitary matrix 
+*  $V$ is a unitary matrix
 *  $V'$ is the adjoint / conjugate transpose of $V$
 "
 
@@ -400,7 +400,7 @@ hint(
 @bind n Slider(1:50, default=50, show_value=true)
 
 # ╔═╡ 8401a066-f5c0-4d86-8d0e-e9c287762252
-md"Note that from a numerical perspective, it is a bit unusual to multiply the factorization $U \tilde{\Sigma} V'$ back together into an array $\tilde{A}$. We did this so we can apply the low-rank approximation to our image. 
+md"Note that from a numerical perspective, it is a bit unusual to multiply the factorization $U \tilde{\Sigma} V'$ back together into an array $\tilde{A}$. We did this so we can apply the low-rank approximation to our image.
 
 Let's compute a rank $n=$ $(n) approximation. Adjust the rank using the following slider:"
 
@@ -436,7 +436,7 @@ $k(x_i, x_j) = \exp\left(-\frac{||x_i - x_j|| ^2}{2\sigma^2}\right)$
 )
 
 # ╔═╡ 7f9afd24-fd1c-49cd-b6bf-598cea9758c7
-md"Note that you can type σ using `\sigma<TAB>`. 
+md"Note that you can type σ using `\sigma<TAB>`.
 
 `σ`'s default value `σ_slider` is defined at the bottom of this notebook."
 
@@ -592,7 +592,7 @@ A \ b # Approach B: left division operator
 C \ b # Approach C: left division operator and Cholesky factorization
 
 # ╔═╡ cb3a2a70-7a4c-4adb-9983-95614645971b
-md"In this example, all three approaches compute the correct result. However, matrix inversion is often numerically unstable, which is why it should be avoided. 
+md"In this example, all three approaches compute the correct result. However, matrix inversion is often numerically unstable, which is why it should be avoided.
 
 The approach using the Cholesky factorization is numerically stable, fast and therefore well suited for the next exercise:"
 
@@ -613,14 +613,14 @@ task(
     md"Implement Kernel Ridge Regression in the function `kernel_ridge` below.
 
 ##### Step 1
-Inside the function `kernel_ridge`, construct the training kernel matrix $K_{XX}$ using the function `kernel`. Add a regularization term $\lambda I$, where where $I$ is the identity matrix and $\lambda \in \mathbb{R}$ is a regularization term: 
+Inside the function `kernel_ridge`, construct the training kernel matrix $K_{XX}$ using the function `kernel`. Add a regularization term $\lambda I$, where where $I$ is the identity matrix and $\lambda \in \mathbb{R}$ is a regularization term:
 
 $\tilde{K}_{XX} = K_{XX} + \lambda I$
 
 Don't specify the kernel bandwidth $\sigma$.
 
 ##### Step 2
-Use the Cholesky decomposition on $\tilde{K}_{XX}$ to compute 
+Use the Cholesky decomposition on $\tilde{K}_{XX}$ to compute
 
 $\alpha = (K_{XX} + \lambda I)^{-1}y \quad .$
 
@@ -693,7 +693,7 @@ else
         elseif result isa AbstractArray
             keep_working(
                 md"""
-The `predict` function should return a scalar prediction $\hat{y}$ for scalar inputs $\hat{x}$. 
+The `predict` function should return a scalar prediction $\hat{y}$ for scalar inputs $\hat{x}$.
 Currently, an array of size $(size(result)) is returned.
 
 If the size is (1,) or (1, 1), you can use the function `only` on your output to access the only element in it. Alternatively, use the dot product.
@@ -739,12 +739,12 @@ Assume an input dataset $$X = [x_1, x_2, \ldots x_n] \in \mathbb{R}^{d\times n}$
 *  $n$ is the number of data points in $X$
 
 Substracting the mean over all data points gives us the *mean-centered* data matrix $\hat{X}$.
-	
-**Idea:** PCA with $k$ principal components finds the projection $W_k = [w_1, w_2, \ldots, w_k]$ that maximizes the variance in the projected data $W_k^T\hat{X}$. 
-	
+
+**Idea:** PCA with $k$ principal components finds the projection $W_k = [w_1, w_2, \ldots, w_k]$ that maximizes the variance in the projected data $W_k^T\hat{X}$.
+
 ![PCA Plot](https://i.imgur.com/FEyXwwX.png)
 
-In the example above, this maximization of variance would correspond to a projection $W_1^T\hat{X} = w_1^T\hat{X}$ of all datapoints onto the red arrow, reducing the dimensionality from 2D to 1D. 
+In the example above, this maximization of variance would correspond to a projection $W_1^T\hat{X} = w_1^T\hat{X}$ of all datapoints onto the red arrow, reducing the dimensionality from 2D to 1D.
 "
 
 # ╔═╡ 067c7894-2e3f-43de-998f-4f1a250bc0f7
@@ -778,9 +778,9 @@ Compute the sample covariance matrix $C = \hat{X}\hat{X}^T$.
 
 ##### Step 3
 Compute an eigendecomposition of $C$ into eigenvectors $w_i$ and eigenvalues $\lambda_i$
-using the function `eigen`. 
+using the function `eigen`.
+
 
-	
 Select the eigenvectors $w_i$ corresponding to the $k$ largest eigenvalues $\lambda_i$,
 such that $\lambda_1\ge\lambda_2\ge\ldots\ge\lambda_k$:
 
@@ -815,7 +815,7 @@ md"### Exercise 4.2 – Visualization"
 task(
     md"
 Try to replicate the following plot as on the dataset `X_test` as close as you can.
-	
+
 ![PCA Plot](https://i.imgur.com/FEyXwwX.png)
 
 

homework/H3_Custom_Types.jl

@@ -48,13 +48,13 @@ html"""
 			Homework 3: Custom types
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24<br>
+			TU Berlin, Summer Semester 2024<br>
 		</p>
 	</div>
 """
 
 # ╔═╡ bdcb27c5-0603-49ac-b831-d78c558b31f0
-md"Due date: **Monday, December 11th 2023 at 23:59**"
+md"Due date: **Monday, May 13th 2024 at 23:59**"
 
 # ╔═╡ ddd6e83e-5a0d-4ff0-afe4-dedfc860994c
 md"### Student information"
@@ -685,8 +685,8 @@ Markdown.MD(
         "Task",
         "Optional task (3 extra points)",
         [
-            md"Implement the *Adam* optimizer, which was introduced 
-            in [this paper](https://arxiv.org/abs/1412.6980) by Kingma and Ba. 
+            md"Implement the *Adam* optimizer, which was introduced
+            in [this paper](https://arxiv.org/abs/1412.6980) by Kingma and Ba.
             Add it to the visualization above and compare it to Optimizer.jl's Adam implementation.",
         ],
     ),

homework/H4_Deep_Learning.jl

@@ -48,13 +48,13 @@ html"""
 			Homework 4: Deep Learning
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24<br>
+			TU Berlin, Summer Semester 2024<br>
 		</p>
 	</div>
 """
 
 # ╔═╡ ce8fd3c2-4fe8-4408-8479-efc81f1a4147
-md"Due date: **Monday, December 18th 2023 at 23:59**"
+md"Due date: **Monday, May 20th 2024 at 23:59**"
 
 # ╔═╡ 42a8576f-d0ab-4915-864f-c1c2fd65e4e4
 md"### Student information"
@@ -178,7 +178,7 @@ convert2image(MNIST, x), y
 
 # ╔═╡ efb62b97-c6fc-495f-a0ae-bf0c2e5485d3
 md"#### Preprocessing
-We preprocess the data for use with Flux.jl by one-hot encoding the targets `y` 
+We preprocess the data for use with Flux.jl by one-hot encoding the targets `y`
 and reshaping the features `x` to WHCN (width, height, color channels, batch size) format:"
 
 # ╔═╡ a173003f-9b32-4042-a263-b5af35106587
@@ -209,7 +209,7 @@ md"Run training: $(@bind run_training CheckBox(default=false))"
 # ╔═╡ fe22ab4f-6119-4397-ac25-ce840a176777
 if run_training # Do NOT modify this!
 
-    # Write your code here
+	# Write your code here
 end
 
 # ╔═╡ ae26551a-d4dd-4b37-a02d-a089de359f70
@@ -333,8 +333,8 @@ task(
 # ╔═╡ 0b11ef46-0c80-4255-bb22-23281704916f
 task(
     md"""
-Adapt the training loop from *Lecture 7* 
-to train the LeNet-5 model from exercise 1.2 (with `MyDense` layers) 
+Adapt the training loop from *Lecture 7*
+to train the LeNet-5 model from exercise 1.2 (with `MyDense` layers)
 on the Fashion-MNIST dataset.
 1. Use the `Flux.logitcrossentropy` loss function
 2. Use a suitable gradient-based optimizer, e.g. `Descent` or `Adam`

lectures/E1_Installation.jl

@@ -49,7 +49,7 @@ html"""
 			Adrian Hill
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24
+			TU Berlin, Summer Semester 2024
 		</p>
 	</div>
 """
@@ -188,11 +188,11 @@ You can then copy this URL into the *"Open a notebook"* field of the Pluto start
 
 ![Starting page](https://i.imgur.com/o7TjYX5.png)
 
-Doing this will ensure that you are working with the latest version of the lectures and homework. 
+Doing this will ensure that you are working with the latest version of the lectures and homework.
 """
 
 # ╔═╡ f1d4d38b-5593-4493-976a-e4b9d486007a
-tip(md"If you are familiar with Git, you can also clone the repository of this course. 
+tip(md"If you are familiar with Git, you can also clone the repository of this course.
 
 Just make sure to regularly `git pull` to keep your copy of the course up to date.")
 

lectures/E2_Help.jl

@@ -32,7 +32,7 @@ html"""
 			Adrian Hill
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24
+			TU Berlin, Summer Semester 2024
 		</p>
 	</div>
 """

lectures/L1_Basics_1.jl

@@ -42,7 +42,7 @@ html"""
 			Adrian Hill
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24
+			TU Berlin, Summer Semester 2024
 		</p>
 	</div>
 """
@@ -335,7 +335,7 @@ typeof(π)
 
 # ╔═╡ b41bee43-f510-483f-9a12-d9d577367272
 md"### Rational numbers
-For integers $p$ and $q$, 
+For integers $p$ and $q$,
 rational numbers $\frac{p}{q}$ can be expressed as fractions using the syntax `p//q`.
 "
 

lectures/L2_Basics_2.jl

@@ -56,7 +56,7 @@ html"""
 			Adrian Hill
 		</p>
 		<p style="font-size: 20px;">
-			TU Berlin, Winter Semester 23/24
+			TU Berlin, Summer Semester 2024
 		</p>
 	</div>
 """
@@ -156,7 +156,7 @@ v7 = [sqrt(x) for x in 1:10 if x % 2 == 0]
 
 # ╔═╡ c66a192f-8bd7-462d-a062-cb670b07b723
 md"""## Indexing
-We can index into vectors using square bracket notation. 
+We can index into vectors using square bracket notation.
 
 Let's create a small test vector with values from 1 to 10 for this purpose:
 """
@@ -244,8 +244,8 @@ end
 # ╔═╡ 43c078fb-7964-48e0-92fd-55d306f36b1d
 #! format: off
 tip(
-md"If a method requires a lot of calls to `@view`, 
-you can also use the `@views` macro (with an extra `s`) on a function, loop, or `begin ... end` block 
+md"If a method requires a lot of calls to `@view`,
+you can also use the `@views` macro (with an extra `s`) on a function, loop, or `begin ... end` block
 to turn every indexing operation inside of it into a view!
 
 **Example:**
@@ -254,7 +254,7 @@ The previous code block could have been written as
 ```julia
 @views begin # every indexing operation in this code block will create a view
 	x2 = collect(1:10)
-	y2 = x2[2:4] # creates view onto x2 due to @views 
+	y2 = x2[2:4] # creates view onto x2 due to @views
 	y2[1] = 42
 end
 ```
@@ -312,8 +312,8 @@ typeof(t1)
 # ╔═╡ 79c55150-0457-49b7-90a5-2cad980b2cb9
 tip(
     md"""
-For performance reasons, it is best to avoid Vectors of type `Vector{Any}`: Julia's compiler can't infer element types and therefore also can't specialize code, resulting in bad performance.  
-	
+For performance reasons, it is best to avoid Vectors of type `Vector{Any}`: Julia's compiler can't infer element types and therefore also can't specialize code, resulting in bad performance.
+
 You can find out more about type promotion in the [Julia documentation on Conversion and Promotion](https://docs.julialang.org/en/v1/manual/conversion-and-promotion/#conversion-and-promotion).
 """,
 )
@@ -856,7 +856,7 @@ end
 
 # ╔═╡ b857984b-ec1b-4409-bdf0-438228950f39
 md"""# Broadcasting on arrays
-We've already seen broadcasting on vectors in the previous lecture. 
+We've already seen broadcasting on vectors in the previous lecture.
 For higher-dimensional arrays, the behaviour is [a bit more complicated]((https://docs.julialang.org/en/v1/manual/arrays/#Broadcasting)):
 
 >  Broadcast **expands singleton dimensions in array arguments to match the corresponding dimension in the other array** without using extra memory, and applies the given function elementwise
@@ -868,9 +868,9 @@ P &\in \mathbb{R}^{2×3×1×1×1} \\
 Q &\in \mathbb{R}^{1×1×4×5} \quad .
 \end{align}$
 
-Julia will expand $P$ and $Q$ to 
+Julia will expand $P$ and $Q$ to
 
-$\tilde{P},\,\tilde{Q} \in \mathbb{R}^{2×3×4×5×1}$ 
+$\tilde{P},\,\tilde{Q} \in \mathbb{R}^{2×3×4×5×1}$
 
 
 without allocating extra memory, then add them:
@@ -900,7 +900,7 @@ size(R)
 
 # ╔═╡ b2cdd7d0-e009-4fca-a16b-ddd19cabc6a0
 md"""# Further resources
-In this lecture, we didn't get to cover Cartesian Indices, which are a great tool for implementing algorithms on high-dimensional arrays. 
+In this lecture, we didn't get to cover Cartesian Indices, which are a great tool for implementing algorithms on high-dimensional arrays.
 The following blogpost gives a great introduction:
 
 - [Multidimensional algorithms and iteration](https://julialang.org/blog/2016/02/iteration/)  by Tim Holy