Cagnur TK

Demystifying MLOps: Enhancing Machine Learning Workflow and Addressing Key Challenges

2023-06-30T00:00:00+00:00

What is MLOps?

Main Source:

https://azure.microsoft.com/en-in/resources/drive-efficiency-and-productivity-with-machine-learning-operations/

MLOps:

Concept or way of working. Not a product or service
Approach for streamlining end-to-end ML life cycle
Does for Data scientists and/or ML engineers what DevOps does for developers

What problems does it address?

MLOps addresses the following gaps:

DevOps teams strive to automate all their workflows, and for the most part, today they have. However, data scientists/ml engineers are excluded. For example, Git is a perfect tool for building, testing and deploying to a production environment. However, due to the nature of ML projects, Git can be an environment that makes this problematic.
ML projects may require tuning, scalability, automation, and packaging. These tasks may take months, and the process is iterative. DevOps does not propose a strategy to support this continuity.

Versioning

What is dataset/model versioning?

It is the process of applying different naming of different stages of a data/model.

In which cases would it be useful?

Facilitating easy sharing
Collaboration
Model reuse in an environment that includes comprehensive access control
Traceability

How does it differ from software versioning?

In ML, the data changes over time, and models need to be retrained regularly, which is not the case for software engineering. If the module requires the requirements in the software engineering field, it is frozen.

ML in Academia vs. Industry

What are the differences between doing ML in academic and industrial settings?

	Academia	Industry
Approach to Metric	Reach an unreached value	Reach an unreached value with many constraints such as less complexity, and low computational cost.
Data	Encourage to open source	Confidentiality of both company and customer data
Model	No constrains	Large models are not preferred (as they will be used in real-time)
Deployment	No need deployment	Optimization has to be done considering the deployment phase
Pipeline	No constrains	Long-term and sustainable (Since it has deployment and mass production, it has a long life cycle)
Owner	Individual or Team	Company
Motivation	Internally motivated	Product-oriented
Money Support	Rare or few (at least for Turkey)	Full support from the company

Performance Issues

Say your model has a decent test set performance but performs poorly in production. What could be the issue?

I assume that the performance refers to the metric in this question. Let’s say you get high accuracy in your test set but not in production. These are the possible reasons:

There might be a distribution shift:
- The customer uses different data distribution than yours. For example, your test set does not cover your customer’s need: Bilby is an Australian animal, not found in the animal photos you collected in Turkey. But one of your customers is from Australia and unhappy with this situation.
- The customer uses different data structures. For example, your test set for image segmentation problems has a high resolution, ex. 1080. But the customers have low resolution, ex 360.
There might be a concept drift. (“change in the relationships between input and output data in the underlying problem overtime”)
- Although you consider your customer’s needs and your test set covers perfectly, the concept of that field may change.

If we assume that the performance refers to the needs:

There might be a computational cost issue:
- Your test environment has a higher and more powerful GPU but the production has poor. Although the model fits in the GPU, the evaluation process takes longer in real time. This could be a poor performance from the customer’s point of view.
- The model output requires memory. Your test environment has high memory, and memory is not an issue for you. But the product has less memory. Let us say after a few usages; the customer could not use the product without deleting the old ones because the memory is full. This could be a poor performance from the customer’s point of view.

How would you address this problem?

The data in which the model fails is examined manually. Try to find a pattern among the data with an error. After finding, a dataset should be created and the model should be retrained with this new dataset. (It is for deep learning. If the data is for machine learning, a discovered or missing feature should be added.)

The Power of Modeling: Unlocking the Wonders of Science

2023-06-29T00:00:00+00:00

Back in 2017, during my time as an engineering student, I had the opportunity to visit the Boston Museum of Science. Little did I know that this visit would profoundly impact my understanding of the world around me. The museum’s dedicated section on modeling caught my attention, and I found myself pondering why modeling, something seemingly simple, warranted such a significant portion of the museum.

It wasn’t until later, when I embarked on a Computational Neuroscience Ph.D. course, that I began to unravel the remarkable power of modeling. In this article, I will delve into three key aspects that make modeling an indispensable tool in scientific exploration.

Encouraging us to think about the mechanism

One of the primary strengths of modeling is its ability to prompt us to contemplate the inner workings of a system. By breaking down complex phenomena into simpler components, modeling allows us to investigate the underlying mechanisms more effectively. It provides a transition from a holistic perspective to a detailed one, granting us the opportunity to question the intricacies of the system.

Shifting the focus to the details Modeling permits us to delve into the finer details of a system, enabling a deeper understanding of its intricacies. By asking probing questions about the behavior of individual components, we can gain insights into the overall behavior of the system as a whole. This shift from the big picture to the specifics helps us uncover valuable knowledge that might otherwise remain hidden.
Identifying dependent and independent variables Another advantage of modeling is its capacity to identify the interrelationships between different variables within a system. By isolating the dependent and independent variables, we can make informed assumptions and simplify the modeling process. This analytical framework allows for more accurate predictions and a clearer understanding of cause-and-effect relationships.

Encouraging us to observe the environment

Modeling also acts as a catalyst for enhanced environmental awareness. When engaging in the modeling process, we are prompted to consider whether the mechanisms we are examining are similar to those present in the real-world scenarios we encounter. This prompts a conscious evaluation of the environment and encourages us to connect our models to the complexities of the natural world.

Increasing environmental awareness As we develop models, we naturally become more attuned to our surroundings. We begin to notice patterns and similarities between the mechanisms we study and those occurring in our environment. This heightened awareness fosters a deeper understanding of the world and its interconnectedness, helping us make more informed decisions and solutions.
Cultivating curiosity and questioning The act of modeling nurtures an innate sense of curiosity and an eagerness to question the world around us. As we explore different variables and interactions, we often uncover new mysteries that ignite our curiosity. This intrinsic motivation to seek answers pushes scientific progress forward and fuels our desire to unravel the complexities of the universe.

Providing a foundation for complex mechanisms

Modeling serves as the bedrock upon which numerous scientific disciplines are built. It has played a pivotal role in shaping various fields, enabling breakthroughs that would have otherwise been unattainable.

Hodgkin-Huxley and the birth of Computational Neuroscience The pioneering work of Hodgkin and Huxley in modeling the behavior of neurons marked the birth of computational neuroscience. Their efforts to mathematically describe the electrical activity of neurons laid the foundation for a field that has revolutionized our understanding of the brain and paved the way for advancements in artificial intelligence and neural networks.
Expanding fields through diverse modeling approaches Modeling’s influence extends across a wide range of disciplines, with each unique modeling approach addressing different functionalities and contributing to the development of its respective field. Whether it’s the simulation of physical phenomena, the economic modeling of markets, or the ecological modeling of ecosystems, each application of modeling brings new insights, advances understanding, and propels the growth of diverse scientific domains.

The Boston Museum of Science’s emphasis on modeling sparked my initial curiosity and led me to explore its profound impact. By encouraging us to think deeply about mechanisms, observe our environment, and serve as the foundation for complex systems, modeling empowers us to unravel the mysteries of the world. It fosters a scientific mindset rooted in curiosity and critical thinking, pushing the boundaries of knowledge across various disciplines. As we continue to harness the power of modeling, we unlock new avenues of understanding and pave the way for groundbreaking discoveries that shape our future.

Unraveling Teacher-Student Knowledge Distillation: Enhancing Small Model Performance

2023-06-18T00:00:00+00:00

Sources for answering the questions:

Take a look at the paper

Teacher-student knowledge distillation

The transferred knowledge of the cumbersome/teacher model, which has a higher capacity and requires a massive amount of computation, to distilled/student model, which cannot be fully utilized with the train from scratch with a large dataset. Instead of the student model learning the information in the data directly, called hard target, the teacher model teaches the student the relational information known on the data, called soft target.

Why mimicking a trained large-capacity model would benefit a smaller model?

The smaller model cannot utilize large datasets. On the other hand, the large model can. It can extract information from very large, highly redundant datasets requiring regularization, such as dropout. The knowledge acquired by the large model also includes the relationship between the data. Learning the relationship between the information obtained by the large model from the vast data provides an advantage in terms of the small model’s capacity. Also, the small model has fast inference time due to its small size. It allows us to have a fast inference time by imitating the information of the large model.

Soft Targets & Temperature (in the context of TSKD)

Softmax stresses the larger value; however, it may lose relativeness with other values. For example, the input set {1,2,3} gives approximately {0.09, 0.24, 0.66}. The inputs can be scaled with the temperature value and affect the outputs. To illustrate, for the same input set and T is 5, the output is approximately {0.26, 0.32, 0.40}. It highlights the big one; it does not change. While with the increasing temperature, their relationship became more visible.

The same logic applies to ML. The softmax function is generally used as an output layer activation function since its result carries probability distribution properties. For instance, if the result gives an output with a probability of close to three among ten possible classes, we can deduce the properties of these three classes are similar.

Due to the nature of the Softmax function, it may not be easy to see because of the convergence at the ends (If the values are high). We will apply the temperature hyperparameter to them; we move them to an area where we can see their relationship quickly. In this way, we make the probability distribution between classes smoother.

We apply this concept to the pre-trained teacher model to train the student model. We teach the outputs of the teacher model with a specific temperature value (which is a soft target) to the student model with the same temperature value.

Alternative Approach

The following method is the Knowledge Distillation method, which does not employ the Teacher-Student paradigm to transfer knowledge.

“Self-Knowledge Distillation”. It has two branches according to [3]: Data augmentation-based approaches and the auxiliary branch-based self-distillation.

Importance of MLOps: Two Survey Articles’ Summaries

2023-04-30T00:00:00+00:00

In this section I presents two important review papers’ summary:

Challenges in Deploying Machine Learning: a Survey of Case Studies by Paleyes, Urma, Lawrence
Hidden Technical Debt in Machine Learning Systems by Sculley et al.

Challenges in Deploying Machine Learning: A Survey of Case Studies

Figure 1: Considerations, Issues, and Concerns explored in the study

The study taxonomizes the challenges in the machine learning (ML) deployment stage into five topics. The first four are technical, and the last one is non-technical: Data Management, Model Learning/Selection, Model Verification, Model Deployment, and Cross-Cutting Aspects. Each occupies a place in the deployment phase, and they are processes that can progress by receiving feedback from each other.

Data management consists of data collection, preprocessing, augmentation, and analysis— data collection questions the existence of the data, and data discovery. Although the data/dataset exists, it requires preprocessing, such as data dispersion and cleaning. Data augmentation simply refers to an increasing number of data. However, if the task is not unsupervised, the labeling problem arises. Other concerns here are access to experts, and the available data is not rich enough to cover the entire problem. Analyses of the data cover the studies to improve data quality.

Model learning includes three subtopics: Model training, model selection, and hyper- parameter selection/hyper-parameter optimization (HPO). Computational cost and environmental impacts are the main concerns for model training. The model selection has three issues: If the model is complex, the teams may overwhelm to optimize. If the resource is restricted, the model should not slow down the iterative process. If the model is not understandable in business domain terms, it makes the output difficult to interpret. One issue of HPO is related to resources: Staying away from options that will increase your hyperparameter set is logical if there are limited resources to try the elements in the set. HPO often needs to consider specific requirements imposed by the model’s environment so hardware-aware optimization is another issue in this subtopic.

Model verification challenges constitute three aspects: requirement encoding, formal verification, and test-based verification. Not only reliable performance but also business- driven performance is added to the requirements. The verification requires high-quality standards, which are formally set via extensive regulatory frameworks. Tests should consider the whole real-world environment, which is challenging to be alone. However, simulations can help in this case but are based on some assumptions.

Model deployment challenges focus on the ineffectiveness of DevOps in the machine learning process. The study examines it under three headings: Integration, Monitoring, and Updating. Integration refers to “implementing the model itself in a form that can be consumed and supported.” (Although the table mentions “operational support” under this section, I could not find any specific paragraph for this.)

There are non-technical challenges that ML brings to our life: ethical and security concerns. Gaining the end user’s trust is considered a challenge in this study. Ethical issues are the following: Restricting access to data due to some regulations that the country restricts, engineers only thinking technically and not thinking about the social dimension, the labeling of the trained systems by us, who are prejudiced, a lot of artworks are given for educational purposes for machine learning to produce art content, the issue of who will own the property rights of the output, and, last but not least, empowering decision-making on ethical matters. The upcoming concerns are from security. All of them are adversarial attacks: intentionally poisoning the data used in training, stealing the model using the input and model output given to the model, and reconstructing training data from model parameters, model inversion. The possibility of not using an application that you put on the field with a lot of effort and not giving confidence to the end-user was also discussed with some practical solutions. The possible answers include including the end-user in the process, giving importance to the user’s feedback, and explaining the model to the end user. The study emphasizes that the last proposal should be a model metric: explainability score.

After introducing these challenges, the study classified proposed solutions to these challenges: Tools & services and Holistic approaches. Although the study’s main contributions are to overlook the challenges in practices in ML, the main goal is to show the academic world’s awareness of these challenges.

Hidden Technical Debt in Machine Learning Systems

Figure 2: Add title

The study makes us think that the things that machine learning brings in have a price. It covers technical debts under seven topics: the border erosion -a boundary problem brought by complexity-, dependency, analysis, configuration debts, anti-patterning -problems arising from not having a specific pattern-, instability of the external world, and others.

Firstly, the study discusses boundary erosion. Encapsulation is not possible for machine learning. It mixes the signal, so it is entanglement. Changing Anything Changes Everything (CACE) the rule states itself very clearly. Two mitigation strategies are assembling and detecting as they occur. Solving a different and new problem with minor corrections makes it dependent on the change in the old problem. This kind of chain fix leads to improvement deadlock. Suggested strategies for correction cascades are proceeding by changing the root model, not chaining, or solving a new problem by training a completely different model. The equivalent of visibility debt in machine learning is mentioned under this section as undeclared consumers. However, they will state their subtitles under analysis debt.

Secondly, the dependencies are discussed. Although dependency can be code and data- related, data dependency is the key player because of its untangled characteristics. Unstable data dependencies can happen either implicitly or explicitly. Both are dangerous as an input signal that looks like an improvement can negatively affect another system using an output signal. Creating a versioned copy of a given signal is suggested to prevent this. As we delete the modules that are not necessary for the code, the input signals that do not provide much benefit should also be removed. This situation is called underutilized data dependency in the study, and they also taxonomies these features as legacy, bundled, epsilon, and correlated. For the detection of these features, Leave-one-feature-out evolution is advised. Static data analysis is another aspect of error checking, tracking down consumers, and enforcing migration and updates. An automated feature management system helps to make the migration and deletion much safer in practice.

Thirdly, ML has analysis debt because it has feedback in it. It could be direct or hidden. It can be a direct feedback loop such as active learning. As a strategy, adding randomization or isolation is implied. It can be hidden since the world has indirect since it is no surprise that the two systems affect each other indirectly in the world, and there is not any direct strategy to detect this.

Fourthly, the study debates the anti-patterning of ML. The difficulties of not having a specific pattern as in traditional methods are mentioned. Using a generic package may prevent us from taking advantage of the domain-specific features or deviating the objective function from our domain-oriented purpose, which refers to glue code. Pipeline jungles is a glue code case and refer to the data preprocessing phase. The study stated that the separate work of engineers and researchers is the root reason. Also, the study expresses another factor: dead experiment code paths, which occur as a natural consequence of both. Periodic reexamining of each branch to see what can be removed is a strategy for this problem—all of these highlight the lack of solid abstractions of ML systems, and abstraction debt. There is not any abstraction like the traditional approach. Lastly, the study mentions some “smells” in ML that could be indicators. These are plain-old-data type, multiple-language, and prototype smells.

Then, Configuration debt is taken up. It is mentioned that the calibration is ignored at first, but this leads to a structural error. The study defines six different principles to prevent calibration debt.

As the ML system’s inputs are taken from the external world, the world’s instability is considered a different subtopic. The world consists of dynamic systems however, the model can have fixed thresholds. Instead of this, adaptive thresholds can be used. Real-time monitoring and testing of the system are required for long-term reliability. Of course, the question of what to monitor arises. For this, the study offers three different starting points. It is emphasized in the article that monitoring and testing are essential to developing this mechanism to react in real-time.

Finally, the study mentions other debts: Data testing, reproducibility, process management, and culture. Checking data distribution can be helpful in a robust, well-functioning system, but it is another debt. Reproducibility, a necessity brought by science, is also a particular problem: The randomness in machine learning makes it difficult to follow this rule. The cost of maintaining what we mention so far is another debt, and it has a managerial obligation. Creating an environment where ML engineers and researchers can work together, and programming appropriately brings a cultural debt.

The study aims to support the development of the concept of sustainability of ML projects. Also, it emphasizes that this sustainability can be achieved with the awareness of both ML engineers and ML researchers.

Getting Started

2017-01-01T00:00:00+00:00

This is a portfolio Jekyll theme built from the ground up, using the DevTips Starter Kit as a foundation for starting, and following closely the amazing tutorial by Travis Neilson over at DevTips. For a guide on how to start a blog or personal website using Jekyll and GitHub Pages, please check out this article.

Everything that you will ever need to know about this Jekyll theme is included in the README below, which you can also find in the demo site. For a guide on how to deploy a Jekyll site using GitHub Pages, please check out this article.

If you like my work then please consider supporting me with Ko-fi.

Notable features

Compatible with GitHub Pages.
Support for Jekyll’s built-in Sass/SCSS preprocessor and data files for making customizing easier.
Google Analytics support.
Commenting support powered by Disqus.
Optimized for search engines.
LaTeX support through MathJax.

Introduction
1. What is Jekyll
2. Never Used Jeykll Before?
Installation
Configuration
Features
Everything Else
Contributing
Questions?
Credits
License

Introduction

Portfolio Jekyll Theme is a Jekyll theme that was built to be 100% compatible with GitHub Pages. If you are unfamiliar with GitHub Pages, you can check out their documentation for more information. Jonathan McGlone’s guide on creating and hosting a personal site on GitHub is also a good resource.

What is Jekyll?

Jekyll is a simple, blog-aware, static site generator for personal, project, or organization sites. Basically, Jekyll takes your page content along with template files and produces a complete website. For more information, visit the official Jekyll site for their documentation.

Never Used Jekyll Before?

The beauty of hosting your website on GitHub is that you don’t have to actually have Jekyll installed on your computer. Everything can be done through the GitHub code editor, with minimal knowledge of how to use Jekyll or the command line. All you have to do is add your posts to the _posts directory and edit the _config.yml file to change the site settings. With some rudimentary knowledge of HTML and CSS, you can even modify the site to your liking. This can all be done through the GitHub code editor, which acts like a content management system (CMS).

Installation

GitHub Pages Installation

To start using Portfolio Jekyll Theme right away with GitHub Pages, fork the Portfolio Jekyll Theme repository on GitHub. From there, you can rename your repository to ‘USERNAME.github.io’, where ‘USERNAME’ is your GitHub username, and edit the settings.yml file in the _data folder to your liking. Ensure that you have a branch named gh-pages. Your website should be ready immediately at ‘http://USERNAME.github.io’. Note: if you are hosting several sites under the same GitHub username, then you will have to use Project Pages instead of User Pages - just change the repository name to something other than ‘http://USERNAME.github.io’.

Head over to the _posts directory to view all the posts that are currently on the website, and to see examples of what post files generally look like. You can simply just duplicate the template post and start adding your own content.

Local Installation

For a full local installation of Portfolio Jekyll Theme, download your own copy of Portfolio Jekyll Theme and unzip it into it’s own directory. From there, open up your favorite command line tool, enter bundle install, and then enter jekyll serve. Your site should be up and running locally at http://localhost:4000.

Directory Structure

If you are familiar with Jekyll, then the Portfolio Jekyll Theme directory structure shouldn’t be too difficult to navigate. The following some highlights of the differences you might notice between the default directory structure. More information on what these folders and files do can be found in the Jekyll documentation site.

Portfolio Jekyll Theme/
├── _data                      # Data files
|  └── settings.yml            # Theme settings and custom text
├── _includes                  # Theme includes
├── _layouts                   # Theme layouts (see below for details)
├── _posts                     # Where all your posts will go
├── assets                     # Style sheets and images are found here
|  ├── css                     # Style sheets go here
|  |  └── all.sass             # Main SASS file
|  └── img                     # Images go here
├── pages                      # Menu pages
├── projects                   # Project pages
├── _config.yml                # Site build settings
├── Gemfile                    # Ruby Gemfile for managing Jekyll plugins
└── index.md                   # Home page

Starting From Scratch

To completely start from scratch, simply delete all the files in the _posts, assets/img, and menu folder, and add your own content. You may also replace the README.md file with your own README. Everything in the _data folder and _config.yml file can be edited to suit your needs. You may also change the favicon.ico file to your own favicon.

Configuration

Sample Posts

Visit the the demo site to find sample posts that show what different types of text formatting look like. You can find these posts in the _posts folder, which show what the best practices for setting up your own site are.

Site Variables

To change site build settings, edit the _config.yml file found in the root of your repository, which you can tweak however you like. More information on configuration settings and plugins can be found on the Jekyll documentation site. This is also where you will be able to customize the title, description, and the author/owner of your site.

If you are hosting your site on GitHub Pages, then committing a change to the _config.yml file will force a rebuild of your site with Jekyll. Any changes made should be viewable soon after. If you are hosting your site locally, then you must run jekyll serve again for the changes to take place.

In the settings.yml file found in the _data folder, you will be able to customize your site settings, such as setting Disqus comments, Google Analytics, what shows up in your menu, and social media information.

The menu pages are found in the menu folder in the root directory, and can be added to your menu in the settings.yml file. Similarly, the project pages are found in the projects folder in the root directory, and can be added to your site in the settings.yml file. The images for your portfolio projects are stored in the assets/img/projects/proj-n folder, and should be named thumb.jpg.

Posts

You will find example posts in your _posts directory. Go ahead and edit any post and re-build the site to see your changes. You can rebuild the site in many different ways, but the most common way is to run jekyll serve, which launches a web server and auto-regenerates your site when a file is updated.

To add new posts, simply add a file in the _posts directory that follows the convention of YYYY-MM-DD-name-of-post.md and includes the necessary front matter. Take a look at any sample post to get an idea about how it works. If you already have a website built with Jekyll, simply copy over your posts to migrate to Portfolio Jekyll Theme.

Layouts

There are two main layout options that are included with Portfolio Jekyll Theme: post and page. Layouts are specified through the YAML front block matter. Any file that contains a YAML front block matter will be processed by Jekyll. For example:

---
layout: post
title: "Example Post"
---

Examples of what posts looks like can be found in the _posts directory, which includes this post you are reading right now. Posts are the basic blog post layout, which includes a header image, post content, author name, date published, social media sharing links, and related posts.

Pages are essentially the post layout without any of the extra features of the posts layout. An example of what pages look like can be found at the About and Contacts.

In addition to the two main layout options above, there is also a custom layout that have been created for the home page. These are simply just page layouts with some Liquid template code. Check out the index.html file in the root directory for what the code looks like.

YAML Front Block Matter

The recommended YAML front block is:

---
layout:
title:
---

layout specifies which layout to use, and title is the page or post title. Have a look at some posts in the _posts directory to see how these variables are set.

Features

Design Considerations

Portfolio Jekyll Theme was designed to be a minimalist theme in order for the focus to remain on your content. For example, links are signified mainly through coloured font, in order to maximize the perceived affordance of clickability (I originally just wanted to make the links a darker shade of grey).

Disqus

Portfolio Jekyll Theme supports comments at the end of posts through Disqus. In order to activate Disqus commenting, set disqus.comments to true in the _data/settings.yml file. If you do not have a Disqus account already, you will have to set one up, and create a profile for your website. You will be given a disqus_shortname that will be used to generate the appropriate comments sections for your site. More information on how to set up Disqus.

Google Analytics

It is possible to track your site statistics through Google Analytics. Similar to Disqus, you will have to create an account for Google Analytics, and enter the correct Google ID for your site under google-ID in the settings.yml file. More information on how to set up Google Analytics. Note: If you are not using Google Analytics, please change google-ID to an empty string.

RSS Feeds

Atom is supported by default through jekyll-feed. With jekyll-feed, you can set configuration variables such as ‘title’, ‘description’, and ‘author’, in the _config.yml file.

All social media icons are courtesy of Font Awesome. You can change which icons appear, as well as the account that they link to, in the settings.yml file in the _data folder.

MathJax

Portfolio Jekyll Theme comes out of the box with MathJax, which allows you to display mathematical equations in your posts through the use of LaTeX.

Syntax Highlighting

Portfolio Jekyll Theme provides syntax highlighting through fenced code blocks. Syntax highlighting allows you to display source code in different colors and fonts depending on what programming language is being displayed. You can find the full list of supported programming languages here. Another option is to embed your code through Gist.

Markdown

As always, Jekyll offers support for GitHub Flavored Markdown, which allows you to format your posts using the Markdown syntax. Examples of these text formatting features can be seen below. You can find this post in the _posts directory as well as the README.md file.

Everything Else

Check out the Jekyll docs for more info on how to get the most out of Jekyll. File all bugs/feature requests at Jekyll’s GitHub repo. If you have questions, you can ask them on Jekyll Talk.

Contributing

If you would like to make a feature request, or report a bug or typo in the documentation, then please submit a GitHub issue. If you would like to make a contribution, then feel free to submit a pull request - as a bonus, I will credit all contributors below! If this is your first pull request, it may be helpful to read up on the GitHub Flow first.

Portfolio Jekyll Theme has been designed as a base for users to customize and fit to their own unique needs. Please keep this in mind when requesting features and/or submitting pull requests. Some examples of changes that I would love to see are things that would make the site easier to use, or better ways of doing things. Please avoid changes that do not benefit the majority of users.

Questions?

This theme is completely free and open source software. You may use it however you want, as it is distributed under the MIT License. If you are having any problems, any questions or suggestions, feel free to tweet at me, or file a GitHub issue.

Credits

Creator

Paul Le

Contributors

Be the first to contribute!

Icons + Demo Images

Other

License

Open sourced under the MIT license.

Text Formatting Examples

2014-01-01T00:00:00+00:00

Markdown Support

Basic Formatting

With Markdown, it is possible to emphasize words by making them italicized, using astericks or underscores, or making them bold, using double astericks or double underscores. Of course, you can combine those two formats, with both bold and italicized text, using any combination of the above syntax. You can also add a strikethrough to text using a ~~double tilde~~.

Paragraphs

This is what a paragraph looks like. For the purpose of demonstration, the rest of this paragraph and the next paragraph after will mean absolutely nothing. Proin eget nibh a massa vestibulum pretium. Suspendisse eu nisl a ante aliquet bibendum quis a nunc. Praesent varius interdum vehicula. Aenean risus libero, placerat at vestibulum eget, ultricies eu enim. Praesent nulla tortor, malesuada adipiscing adipiscing sollicitudin, adipiscing eget est. Praesent nulla tortor, malesuada adipiscing adipiscing sollicitudin, adipiscing eget est.

Proin eget nibh a massa vestibulum pretium. Suspendisse eu nisl a ante aliquet bibendum quis a nunc. Mauris lobortis nulla et felis ullamcorper bibendum. Phasellus et hendrerit mauris. Proin eget nibh a massa vestibulum pretium. Suspendisse eu nisl a ante aliquet bibendum quis a nunc. Praesent varius interdum vehicula. Aenean risus libero, placerat at vestibulum eget, ultricies eu enim. Praesent nulla tortor, malesuada adipiscing adipiscing sollicitudin, adipiscing eget est.

Headings

Sometimes it is useful to have different levels of headings to structure your documents. Start lines with # to create headings. Multiple ## in a row denote smaller heading size. The following demonstrate the full range of heading sizes:

Heading One (h1)

Heading Two (h2)

Heading Three (h3)

Heading Four (h4)

Heading Five (h5)

Heading Six (h6)

Links

You can create an inline link by wrapping link text in square brackets [ ], and then wrapping the URL in parentheses ( ). For example, it is very easy to link to Google!.

Blockquotes

Blockquotes are useful for denoting quotes, or highlighting a large block of text. Single line blockquote:

This quote will change your life.

Multi line blockquote with a cite reference:

People think focus means saying yes to the thing you’ve got to focus on. But that’s not what it means at all. It means saying no to the hundred other good ideas that there are. You have to pick carefully. I’m actually as proud of the things we haven’t done as the things I have done. Innovation is saying no to 1,000 things.

Code and Syntax Highlighting

Code blocks are part of the Markdown spec, but syntax highlighting isn’t. However, many renderers - like GitHub or most Jekyll themes - support syntax highlighting. Which languages are supported and how those language names should be written will vary from renderer to renderer. You can find the full list of supported programming languages here. Also, it is possible to do inline code blocks, by wrapping the text in ` ` ` quotations.

No language indicated, so no syntax highlighting.

def print_hi(name)
  puts "Hi, #{name}"
end
print_hi('Tom')
#=> prints 'Hi, Tom' to STDOUT.

// Example can be run directly in your JavaScript console

// Create a function that takes two arguments and returns the sum of those arguments
var adder = new Function("a", "b", "return a + b");

// Call the function
adder(2, 6);
// > 8

Another option is to embed your code through Gist.

Unordered and Numbered Lists

You can make an unordered and nested list by preceding one or more lines of text with -, *, or +, and indenting sublists. The following lists show the full range of possible list formats.

List item one
- List item one
  - List item one
  - List item two
  - List item three
  - List item four
- List item two
- List item three
- List item four
List item two
List item three
List item four

Numbered lists are made by using numbers instead of bullet points.

List item one
1. List item one
  1. List item one
  2. List item two
  3. List item three
  4. List item four
2. List item two
3. List item three
4. List item four
List item two
List item three
List item four

MathJax Example

The Schrödinger equation is a partial differential equation that describes how the quantum state of a quantum system changes with time:

\[i\hbar\frac{\partial}{\partial t} \Psi(\mathbf{r},t) = \left [ \frac{-\hbar^2}{2\mu}\nabla^2 + V(\mathbf{r},t)\right ] \Psi(\mathbf{r},t)\]

Joseph-Louis Millennial was an Italian mathematician and astronomer who was responsible for the formulation of Lagrangian mechanics, which is a reformulation of Newtonian mechanics.

\[\frac{\mathrm{d}}{\mathrm{d}t} \left ( \frac {\partial L}{\partial \dot{q}_j} \right ) = \frac {\partial L}{\partial q_j}\]

Tables

Title 1	Title 2	Title 3	Title 4
lorem	lorem ipsum	lorem ipsum dolor	lorem ipsum dolor sit
lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit
lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit
lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit	lorem ipsum dolor sit

Embedding

Plenty of social media sites offer the option of embedding certain parts of their site on your own site, such as YouTube and Twitter:

New Collection

Inline HTML elements

HTML defines a long list of available inline tags, which you can mix with Markdown if you like. A complete list of which can be found on the Mozilla Developer Network.

Useful Resources

More information on Markdown can be found at the following links:

More Jekyll Themes!

2013-01-01T00:00:00+00:00

Lagrange

Lagrange is a minimalist Jekyll blog theme that I built from scratch. The purpose of this theme is to provide a simple, clean, content-focused blogging platform for your personal site or blog.

Feel free to check out the demo, where you’ll also find instructions on how to use install and use the theme.

Millennial

Millennial is a minimalist Jekyll blog theme that I built from scratch. The purpose of this theme is to provide a simple, clean, content-focused publishing platform for a publication or blog.