Advent of Code 2022 in pure TensorFlow - Day 7


Solving problem 7 of the AoC 2022 in pure TensorFlow allows us to understand certain limitations of the framework. This problem requires a lot of string manipulation, and TensorFlow (especially in graph mode) is not only not easy to use when working with this data type, but also it has a set of limitations I’ll present in the article. Additionally, the strings to work with in problem 7 are (Unix) paths. TensorFlow has zero support for working with paths, and thus for simplifying a part of the solution, I resorted to the pathlib Python module, thus not designing a completely pure TensorFlow solution.

Day 7: No Space Left On Device

You can click on the title above to read the full text of the puzzle. The TLDR version is: we are given a terminal output, containing some commands and the results of the execution of these commands. The commands are two standard Unix commands:

  • cd: for changing directory, with support for full paths and relative paths
  • ls: for listing the content of the directory I’m currently in.

The ls output contains information about the size of the type of the various files inside of the current working directory.

So, given a puzzle input like

$ cd /
$ ls
dir a
14848514 b.txt
8504156 c.dat
dir d
$ cd a
$ ls
dir e
29116 f
2557 g
62596 h.lst
$ cd e
$ ls
584 i
$ cd ..
$ cd ..
$ cd d
$ ls
4060174 j
8033020 d.log
5626152 d.ext
7214296 k

the goal is to organize the information and determine the filesystem structure, as presented below.

- / (dir)
  - a (dir)
    - e (dir)
      - i (file, size=584)
    - f (file, size=29116)
    - g (file, size=2557)
    - h.lst (file, size=62596)
  - b.txt (file, size=14848514)
  - c.dat (file, size=8504156)
  - d (dir)
    - j (file, size=4060174)
    - d.log (file, size=8033020)
    - d.ext (file, size=5626152)
    - k (file, size=7214296)

After doing that, we should be able to answer the part 1 question: “Find all of the directories with a total size of at most 100000. What is the sum of the total sizes of those directories?”.

Design Phase

The problem breakdown follows:

  1. Parse the input: differentiate commands execution and commands output
  2. Identify the “current working directory” and sum all the file sizes. Being paths, we should keep track of the absolute path every time we work on a folder, since it may be possible to have folders with the same name in different absolute positions.
  3. Create the mapping absolute path:size for every directory.
  4. Filter the mapped directories, searching for all the directories that satisfy the condition and summing up the sizes for getting the part 1 solution

Input parsing: tf.data.Dataset.scan

tf.data.Dataset.scan is a transformation that scans a function across an input dataset. It’s the stateful counterpart of tf.data.Dataset.map. The state is carried on on every iteration through the transformation scan_func(old_state_input_element) -> (new_state, output_element) via the state tensors.

The idea is to use the scan method for:

  1. Parse the lines. Understand when the line is a command execution or the command output.
  2. Whenever we change directory (command cd), create a new state that contains the “full path” we’ll be visiting on the next iteration, together with the initial size (0). Thus, the initial state for every new folder will be path 0. We use a string so we can split the state, and extract the path and the size.
  3. Every time we encounter something that’s not a command (thus, is a command output) we should understand if it’s a file (because a directory has no size). If it’s a file, parse its size and carry on the state for this path, summing the filesize (e.g. at step i the state is /this/path 10. Then we encounter a file with size 100 at step i+1, thus the state now becomes /this/path 110).
  4. The scan method produces an output value on every iteration, thus we can produce the state as the output value, but only when we change state (so we computed the complete size of a folder). By default, thus, we can use the empty string as output (so later on we can filter the output after applying the scan function) and produce an output in the format path size only when the path has been analyzed
  5. The input contains different cd invocations. We can change directory using absolute or relative paths. Thus, we need to take care of this. Doing it with TensorFlow inside the scan method hasn’t been possible - at least, I haven’t found an easy solution. So, for now, we’ll build an absolute path every time without considering the changes to relative locations. E.g. cd /a/b/c followed by cd .. will become /a/b/c/../ instead of /a/b/. Both paths represent the same absolute location, thus we’ll take care of this later on.

To correctly implement the algorithm described above, we need to add a “fake line” to the dataset, with a cd command, since we trigger the output of the accumulated state when we change directory. Moreover, we need to define a meaningful initial state. We can assume that the input always starts from the root directory /.

dataset = dataset.concatenate(tf.data.Dataset.from_tensors(tf.constant("$ cd /")))
initial_state = tf.constant("/ 0")

Here’s the implementation of the scan_func function that we’ll use the loop over the dataset and extract the pairs /x/y/z/../../i size.

Take your time to read the implementation, it’s not trivial.

def func(old_state, line):
    is_command = tf.strings.regex_full_match(line, r"^\$.*")
    new_state = old_state
    if is_command:
        if tf.strings.regex_full_match(line, r"\$ cd .*"):
            dest = tf.strings.split([line], " ")[0][-1]
            if tf.equal(dest, "/"):
                new_state = tf.constant("/ 0")
            else:
                old_path = tf.strings.split([old_state], " ")[0][0]
                new_state = tf.strings.join(
                    [tf.strings.join([old_path, dest], "/"), "0"], " "
                )
    else:
        split = tf.strings.split([line], " ")[0]
        if tf.not_equal(split[0], "dir"):
            size = tf.strings.to_number(split[0], tf.int64)
            state_size = tf.strings.split([old_state], " ")[0]
            if tf.equal(tf.shape(state_size, tf.int64)[0], 1):
                old_size = tf.constant(0, tf.int64)
            else:
                old_size = tf.strings.to_number(state_size[1], tf.int64)

            partial_size = size + old_size
            new_state = tf.strings.join(
                [
                    tf.strings.split(old_state, " ")[0],
                    tf.strings.as_string(partial_size),
                ],
                " ",
            )

    if tf.not_equal(new_state, old_state):
        output_value = new_state
    else:
        output_value = tf.constant("")

    return new_state, output_value

intermediate_dataset = dataset.scan(initial_state, func)

Once again tf.strings and regex are the tools used for parsing the input and understand (thanks to the regular expression) whether we are parsing a command or not.

Iterating over intermediate_dataset we can get a list of elements such as

['', '/ 14848514', '/ 23352670', ...., '//a/e/../../d 17719346', '//a/e/../../d 24933642', '/ 0']

This scan-based solution created a new state every time a directory has been changed. However, the output produced while looping over the scan-generated dataset is the partial value computed while traversing a directory. For example, the first value of the list is 14848514 that’s precisely b.txt (file, size=14848514). The second value for / is 23352670 that’s b.txt + c.dat.

Thus, we’ll need to filter these values later on, after having resolved the absolute paths (e.g. going from //a/e/../../d to /d).

Resolving paths: not pure TensorFlow solution

Removing empty lines from the intermediate_dataset is trivial and can be done by using the tf.data.Dataset.filter method.

filtered_dataset = intermediate_dataset.filter(
    lambda line: tf.strings.regex_full_match(line, "^.* \d*$")
).map(lambda line: tf.strings.regex_replace(line, r"\/\/", "/"))

In this way, we just filtered all the empty lines and kept only the lines ending with a number. Moreover, we also replaced all the double slashes with a single slash, using a simple regex.

However, the very tough problem to solve is designing a pure TensorFlow solution for resolving all the relative paths. I haven’t been able to design such a solution, and thus I had to use the Python pathlib module, unfortunately. It looks like cheating since this is not a pure TensorFlow solution anymore :(


def gen(ds):
    def resolve():
        for pair in ds:
            path, count = tf.strings.split([pair], " ")[0]
            path = Path(path.numpy().decode("utf-8")).resolve().as_posix()
            yield path, count.numpy().decode("utf-8")

    return resolve

filtered_dataset = tf.data.Dataset.from_generator(
    gen(filtered_dataset), tf.string, output_shapes=[2]
)

Apart from this small cheat, so far so good. The filtered_dataset now contains absolute paths (repeated) and the value computed while traversing that path. So, as anticipated, we need to filter these values computed while traversing the path. They are sequential, so we can once again use the tf.data.Dataset.scan method to change-state every time the path changes, and produce only valid values.

def mapper(old_state, pair):
    old_path = old_state[0]
    new_path = pair[0]
    output_value = tf.constant(["", ""])
    if tf.logical_or(
        tf.equal(old_path, "fake_path"), tf.equal(new_path, "fake_path")
    ):
        output_value = tf.constant(["", ""])
    elif tf.not_equal(old_path, new_path):
        output_value = old_state

    return pair, output_value

initial_state = tf.constant(["fake_path", "-1"])
filtered_dataset = (
    filtered_dataset.concatenate(tf.data.Dataset.from_tensors(initial_state))
    .scan(initial_state, mapper)
    .filter(
        lambda pair: tf.logical_and(
            tf.greater(tf.strings.length(pair[0]), 0), tf.not_equal(pair[1], "0")
        )
    )
)

filtered_dataset will now produce only the pairs containing the full path of each folder and its size, without considering the subfolders.

  • ('/', '23352670')
  • ('/a', '94269')
  • ('/a/e', '584')
  • ('/d', '24933642')

We now need to map the full paths with their corresponding full-size. In this way, we can easily answer the puzzle question.

MutableHashTable: mapping paths to their size

We can, once again, use a Mutable hashmap to iteratively create the correspondence between the full path and its size, also considering the sub-folders.

The algorithm is straightforward. We loop over the filtered_dataset and verify if the path is present. If it’s not present, we insert the key-value pair path,size in the mutable hashmap. Otherwise, we split the path using the directory separator (/) and iteratively add the value to all the parent paths already inserted in the hashmap.

lut = tf.lookup.experimental.MutableHashTable(tf.string, tf.int64, default_value=0)
for pair in filtered_dataset:
    path, value = pair[0], tf.strings.to_number(pair[1], tf.int64)
    parts = tf.strings.split(path, "/")

    if tf.logical_and(tf.equal(parts[0], parts[1]), tf.equal(parts[0], "")):
        keys = ["/"]
        old = lut.lookup(keys)[0]
        new = old + value
        lut.insert(keys, [new])
    else:
        for idx, part in enumerate(parts):
            if tf.equal(part, ""):
                keys = ["/"]
            else:
                l = [tf.constant("")] + parts[1 : idx + 1]
                j = tf.strings.join(l, "/")
                keys = [j]
            old = lut.lookup(keys)[0]
            new = old + value
            lut.insert(keys, [new])

The lut variable now contains all the information we need to solve the problem. The keys of the hashmap are all the available paths, the values are the full-size (considering subfolders) of every path.

Thus, we can answer the puzzle question in a few lines:

paths, sizes = lut.export()
print(paths, sizes)
tf.print(
    "part 1: ",
    tf.reduce_sum(tf.gather(sizes, tf.where(tf.math.less_equal(sizes, 100000)))),
)

Problem solved! Let’s go straight to part 2.

Day 7: No Space Left On Device: part 2

Luckily enough, part 2 presents us with an easy-to-solve problem. The puzzle tells us that our total disk space available is 70000000, and we need to install an update whose size is 30000000, but we don’t have enough space for installing it! The puzzle tells us to find the smallest directory that, if deleted, would free up enough space on the filesystem and print the total size of that directory.

In our lookup table (the mutable hashtable) we have everything that’s needed to solve the problem. We need to:

  1. Find the required space as the difference between the update size and the free space already present in the filesystem
  2. Find all the folders whose size is “big enough” to contain the update (that means, to be deleted to install the update).
  3. Get the smallest folder among the ones that satisfy the above criteria.

The code is a 1:1 mapping with these points.

update_size = 30000000
free_space = 70000000 - lut.lookup("/")
required_space = update_size - free_space

big_enough = tf.gather(
    sizes, tf.where(tf.math.greater_equal(sizes - required_space, 0))
)
tf.print("part 2: ", tf.gather(big_enough, tf.math.argmin(big_enough, axis=0)))

Here we go! Day’s 7 problem solved!

Conclusion

You can see the complete solution in folder 7 in the dedicated GitHub repository (in the 2022 folder): https://github.com/galeone/tf-aoc.

Solving problem 7 demonstrated all the weaknesses of the framework while working with strings and with paths in particular. There are no utilities for working with paths in graph mode, and thus this solution ended-up to use a pathlib utility for resolving the relative paths to their full paths counterparts. In general, tf.data.Dataset.scan has demonstrated to be a powerful method for keeping track of a state while iterating over a dataset, although it’s not perfectly natural to solve problems like that in this way. Nonetheless, it’s a fun exercise.

If you missed the article about the previous days’ solutions, here’s a handy list

For any feedback or comment, please use the Disqus form below - thanks!

Don't you want to miss the next article? Do you want to be kept updated?
Subscribe to the newsletter!

Related Posts

Building a RAG for tabular data in Go with PostgreSQL & Gemini

In this article we explore how to combine a large language model (LLM) with a relational database to allow users to ask questions about their data in a natural way. It demonstrates a Retrieval-Augmented Generation (RAG) system built with Go that utilizes PostgreSQL and pgvector for data storage and retrieval. The provided code showcases the core functionalities. This is an overview of how the "chat with your data" feature of fitsleepinsights.app is being developed.

Using Gemini in a Go application: limits and details

This article explores using Gemini within Go applications via Vertex AI. We'll delve into the limitations encountered, including the model's context window size and regional restrictions. We'll also explore various methods for feeding data to Gemini, highlighting the challenges faced due to these limitations. Finally, we'll briefly introduce RAG (Retrieval-Augmented Generation) as a potential solution, but leave its implementation details for future exploration.

Custom model training & deployment on Google Cloud using Vertex AI in Go

This article shows a different approach to solving the same problem presented in the article AutoML pipeline for tabular data on VertexAI in Go. This time, instead of relying on AutoML we will define the model and the training job ourselves. This is a more advanced usage that allows the experienced machine learning practitioner to have full control on the pipeline from the model definition to the hardware to use for training and deploying. At the end of the article, we will also see how to use the deployed model. All of this, in Go and with the help of Python and Docker for the custom training job definition.

Integrating third-party libraries as Unreal Engine plugins: solving the ABI compatibility issues on Linux when the source code is available

In this article, we will discuss the challenges and potential issues that may arise during the integration process of a third-party library when the source code is available. It will provide guidance on how to handle the compilation and linking of the third-party library, manage dependencies, and resolve compatibility issues. We'll realize a plugin for redis plus plus as a real use case scenario, and we'll see how tough can it be to correctly compile the library for Unreal Engine - we'll solve every problem step by step.

AutoML pipeline for tabular data on VertexAI in Go

In this article, we delve into the development and deployment of tabular models using VertexAI and AutoML with Go, showcasing the actual Go code and sharing insights gained through trial & error and extensive Google research to overcome documentation limitations.

Advent of Code 2022 in pure TensorFlow - Day 12

Solving problem 12 of the AoC 2022 in pure TensorFlow is a great exercise in graph theory and more specifically in using the Breadth-First Search (BFS) algorithm. This problem requires working with a grid of characters representing a graph, and the BFS algorithm allows us to traverse the graph in the most efficient way to solve the problem.

Advent of Code 2022 in pure TensorFlow - Day 11

In this article, we'll show how to solve problem 11 from the Advent of Code 2022 (AoC 2022) using TensorFlow. We'll first introduce the problem and then provide a detailed explanation of our TensorFlow solution. The problem at hand revolves around the interactions of multiple monkeys inspecting items, making decisions based on their worry levels, and following a set of rules.

Advent of Code 2022 in pure TensorFlow - Day 10

Solving problem 10 of the AoC 2022 in pure TensorFlow is an interesting challenge. This problem involves simulating a clock signal with varying frequencies and tracking the state of a signal-strength variable. TensorFlow's ability to handle complex data manipulations, control structures, and its @tf.function decorator for efficient execution makes it a fitting choice for tackling this problem. By utilizing TensorFlow's features such as Dataset transformations, efficient filtering, and tensor operations, we can create a clean and efficient solution to this intriguing puzzle.

Advent of Code 2022 in pure TensorFlow - Day 9

In this article, we'll show two different solutions to the Advent of Code 2022 day 9 problem. Both of them are purely TensorFlow solutions. The first one, more traditional, just implement a solution algorithm using only TensorFlow's primitive operations - of course, due to some TensorFlow limitations this solution will contain some details worth reading (e.g. using a pairing function for being able to use n-dimensional tf.Tensor as keys for a mutable hashmap). The second one, instead, demonstrates how a different interpretation of the problem paves the way to completely different solutions. In particular, this solution is Keras based and uses a multi-layer convolutional model for modeling the rope movements.

Advent of Code 2022 in pure TensorFlow - Day 8

Solving problem 8 of the AoC 2022 in pure TensorFlow is straightforward. After all, this problem requires working on a bi-dimensional grid and evaluating conditions by rows or columns. TensorFlow is perfectly suited for this kind of task thanks to its native support for reduction operators (tf.reduce) which are the natural choice for solving problems of this type.