Project Overview

Try out Will It Fit here!

Project Description

Will It Fit? starts as a group project for the Le Wagon Berlin Data Science Bootcamp, batch 606. The initial pitch can be found here. For the purposes of creating a minimum working product, the following components will be completed first (see more details below):

• User interface
• Web scraping module for IKEA package dimensions
• Volume optimization algorithm
• 3D interactive plot

After the minimum working product is complete, the above components will be refined further, both for speed and versatility. In addition, other potential components include (non-exhaustive list):

• Web scraping module (text and images) for car dimensions
• Machine Learning module to estimate car trunk size
• User interface enhancement allowing user to take photo of own vehicle (as opposed to manually inputing make and model)
• Inclusion of providers other than IKEA

Links

• Trello Board

Team Members

• Katarzyna Cieslik
• Tzu-Fan Tang
• Dominik Wagner
• Florian Reichel

Module Description

User interface (frontend/frontend.py)

• Streamlit interface allowing the following user interactions:
  • User provides list and count of articles.
  • User selects from a list of pre-defined cars (whose trunk size is known).
  • Interface returns visual representation of how packages would fit in trunk after performing the following steps:
    • Parse PDF or article list into dictionary.
    • Obtain trunk volume array from car model database.
    • Pass article dict to IKEA website, get package dimensions for each article in a list returned.
    • Pass article dict and trunk space to optimizer, get filled trunk space and list of package coordinates returned.
    • Pass filled trunk space and list of package coordinates to plotter, receive plot.

Inputs

• By user:
  • Choose IKEA website language/location (important for country-specific IKEA scraping as well as PDF parsing)
  • Car make and model from selection boxes
  • Upload user's IKEA wishlist pdf or specify article codes and counts
  • Choose if back-seat is to be moved for extra space

Outputs

• Visual feedback of every step in the process, Car Image when a specific model is chosen (if it exists in the database), including error messages and computational steps.
• 3D interactive plot in user interface.

Potential further enhancements

• Article data collection:
  • Log into IKEA website using user login details and extract shopping basket data directly.
• Language selection:
  • Offer interface in user language.

IKEA Web Scraper (scrapers/IKEA.py)

• Receives dict of article codes and article counts.
• For existing articles:
  • Obtain relevant data from database.
• For new articles:
  • Scrapes IKEA website to obtain package dimensions (rounded up to next cm), weights (in kg rounded up 2 decimals) and counts.
  • Exports new articles into database.
• Returns list of package dimensions and weights.

Inputs

• Article list and counts: {article_code (str): item_count (int)} (dict)
• Country domain: IKEA_COUNTRY_DOMAIN (str)
• Website language: IKEA_WEBSITE_LANGUAGE (str)

Outputs

• List of package dimensions and weights:

[(
    article_code (str),
    item_count (int),
    [(
        package_id (int),
        package_length (int),
        package_width (int),
        package_height (int),
        package_weight (float)
    )]
)] (list)

Potential further enhancements

• Ability to scrape other countries' websites.
• Inch/cm and pound/kg conversions.
• Also return URL to article as well as direct URL to one picture of article.
• Non-cuboid packages.

Volume Optimization Algorithm (optimizers/volumeoptimizer.py)

• Receives list of package dimensions, weights and counts
• Receives available volume
• Optimizes stacking of packages in available volume, one thread per configuration
• Returns 3D numeric representation of occupied space as well as article coordinates

Inputs

• List of package dimensions and weights:

[(
    article_code (str),
    item_count (int),
    [(
        package_id (int),
        package_length (int),
        package_width (int),
        package_height (int),
        package_weight (float)
    )]
)] (list)

• Available volume: 3-dimensional numpy array, each element is 1cm (see params.py)

Outputs

• Filled volume: 3-dimensional numpy array, each element is 1cm (see params.py)
• Article coordinates:

[
    article_code (str),
    article_id (int),
    package_id (int)
    x_start (int),
    y_start (int),
    z_start (int),
    x_end (int),
    y_end (int),
    z_end (int)
] (list)

Potential further enhancements

• Train a machine learning model on a set of already-optimized configurations, eliminating the requirement for individual further optimizations.
• Include weight as a factor - heavier items should sit near the bottom.

3D Interactive Plot (plotting/plotter.py)

• Receives 3D numeric representation of occupied space as well as article coordinates
• Returns interactive 3D plot of packages

Inputs

• Filled volume: 3-dimensional numpy array, each element is 1cm (see params.py)
• Article coordinates:

[
    article_code (str),
    article_id (int),
    package_id (int),
    x_start (int),
    y_start (int),
    z_start (int),
    x_end (int),
    y_end (int),
    z_end (int)
] (list)

Outputs

• 3D interactive plot which can be displayed through user interface

Potential further enhancements

• Add hover labels to packages.
• Add list next to plot. When user hovers over/clicks on article its location on plot is highlighted.
• Hover labels - include article name, link to article and small preview picture (would have to be buffered).

Relevant Parameters (params.py)

These are parameters which can be set before project deployment. Ideally, most relevant parameters should be modifiable from the user interface:

• BIAS_STACKS (list): Controls how the optimizer will try to place the next package and takes two arguments per list element: biased (bool) and bias_tendency (float 0-1). When biased is set to False, the optimizer will choose an entirely random package orientation. When it is set to True, it will, up to bias_tendency, use the orientation that stacks the package as flatly as possible, i.e. the two largest dimensions will be in the original plane. Each list element creates a separate thread.
• GEN_SORTERS (list): Lists pre-defined package sorting orders in the format criterion|direction. So far only volume has been implemented as a criterion. Each sorting order is optimized in a separate thread.
• RANDOM_LIST_COUNT (int): Number of randomized package lists the optimizer should also consider. Each is run in a separate thread. This can be useful to find the optimal stacking solution, but it increases processing time and memory requirements.
• OPT_MAX_ATTEMPTS (int): Number of times each optimizer thread will try to place packages if an individual attempt doesn't succeed, i.e. if the next package cannot be placed anymore. As there is an element of randomness to individual package placement (other than for a biased = True, bias_tendency = 1 package), trying again may yield a different result. This also increases processing time.

Useful Commands

• make install_requirements resets Python virtual environment to only the packages defined in requirements.txt. It removes all other packages.
• make start_app starts the front-end on specified port.
• make docker_build img=IMAGE_NAME mode=MODE builds a Docker image with the specified IMAGE_NAME. If mode=GC is not specified, the image is built locally, otherwise it's built for GC deployment.
• make docker_run img=IMAGE_NAME runs a Docker image locally with the specified IMAGE_NAME.
• make docker_build_run_deploy img=IMAGE_NAME mode=MODE builds and then runs or deploys (mode=GC) the image with the specified IMAGE_NAME.

Change Log

• 18/06/2021: User interface description updated (proxvision)
• 06/06/2021: Added parameter description (freichel)
• 05/06/2021: Clean-up and added useful commands section (freichel)
• 03/06/2021: User interface section updated (proxvision)
• 02/06/2021: Removed superfluous sections
• 31/05/2021: Added space separators and additional info on optimizer (freichel)
• 30/05/2021: Created initial document (freichel)