Reason: A Shell for Research Papers

• Did I ever read this paper?
• Which OSDI 2021 papers did I read?
• Which ones have the word 'Distributed' in their titles?
• How many papers in 2020 were co-authored by Professor Mosharaf Chowdhury?

Well, ask reason.

How it works

Filter and list papers

$ reason
>> # Show all papers.
>> ls
+----------------------------------------------------------+----------------+---------+------+
|                           title                          |  first author  |  venue  | year |
+============================================================================================+
| Shadowtutor: Distributed Partial Distillation for Mobile | Jae-Won Chung  | ICPP    | 2020 |
| Video DNN Inference                                      |                |         |      |
|----------------------------------------------------------+----------------+---------+------|
| Efficient Memory Disaggregation with Infiniswap          | Juncheng Gu    | NSDI    | 2017 |
|----------------------------------------------------------+----------------+---------+------|
| Refurbish Your Training Data: Reusing Partially          | Gyewon Lee     | ATC     | 2021 |
| Augmented Samples for Faster Deep Neural Network         |                |         |      |
| Training                                                 |                |         |      |
|----------------------------------------------------------+----------------+---------+------|
| Finding Consensus Bugs in Etherium via Multi-transaction | Youngseok Yang | OSDI    | 2021 |
| Differential Fuzzing                                     |                |         |      |
|----------------------------------------------------------+----------------+---------+------|
| Tiresias: A GPU Cluster Manager for Distributed Deep     | Juncheng Gu    | NSDI    | 2019 |
| Learning                                                 |                |         |      |
|----------------------------------------------------------+----------------+---------+------|
| Nimble: Lightweight and Parallel GPU Task Scheduling for | Woosuk Kwon    | NeurIPS | 2020 |
| Deep Learning                                            |                |         |      |
+----------------------------------------------------------+----------------+---------+------+
>> # Filter by 'title'. All these are regexes!
>> ls 'Deep Learning$'
+------------------------------------------------------------+--------------+---------+------+
|                            title                           | first author |  venue  | year |
+============================================================================================+
| Tiresias: A GPU Cluster Manager for Distributed Deep       | Juncheng Gu  | NSDI    | 2019 |
| Learning                                                   |              |         |      |
|------------------------------------------------------------+--------------+---------+------|
| Nimble: Lightweight and Parallel GPU Task Scheduling for   | Woosuk Kwon  | NeurIPS | 2020 |
| Deep Learning                                              |              |         |      |
+------------------------------------------------------------+--------------+---------+------+
>> # You may set default filters with `cd`.
>> # BTW, `cd .`, `cd ..`, `cd -`, and `cd` are supported, too.
>> cd 'Deep Learning$'
>> pwd
title matches 'Deep Learning$'
>> # Default filter are automatically applied.
>> # Infiniswap (NSDI'17) is not shown, because its title doesn't match 'Deep Learning$'.
>> ls at NSDI
+------------------------------------------------------------+--------------+---------+------+
|                            title                           | first author |  venue  | year |
+============================================================================================+
| Tiresias: A GPU Cluster Manager for Distributed Deep       | Juncheng Gu  | NSDI    | 2019 |
| Learning                                                   |              |         |      |
+------------------------------------------------------------+--------------+---------+------+
>> # Delete Tiresias.
>> ls at NSDI | rm
Removed 1 paper.

Import new papers

>> # Import directly from arXiv and USENIX. This will also download paper PDFs.
>> curl https://arxiv.org/abs/2105.11367
+--------------------------------------------------------+--------------+-------+------+
|                          title                         | first author | venue | year |
+======================================================================================+
| FedScale: Benchmarking Model and System Performance of | Fan Lai      | arXiv | 2021 |
| Federated Learning                                     |              |       |      |
+--------------------------------------------------------+--------------+-------+------+
>> curl https://www.usenix.org/conference/nsdi21/presentation/you
+------------------------------------------+--------------+-------+------+
|                   title                  | first author | venue | year |
+========================================================================+
| Ship Compute or Ship Data? Why Not Both? | Jie You      | NSDI  | 2021 |
+------------------------------------------+--------------+-------+------+
>> # Modify paper metadata.
>> ls ship | set as Kayak
>> # Or, import manually.
>> touch 'Batch Normalization: Accelerating Deep Network Training by Reducing Internal
Covariate Shift' by 'Sergey Ioffe, Christian Szegedy' at ICML in 2015 as BN @ BatchNorm.pdf
+--------------------------------------------------------------+--------------+-------+------+
|                             title                            | first author | venue | year |
+============================================================================================+
| Batch Normalization: Accelerating Deep Network Training by   | Sergey Ioffe | ICML  | 2015 |
| Reducing Internal Covariate Shift                            |              |       |      |
+--------------------------------------------------------------+--------------+-------+------+

Read, take notes, and create books!

>> # Open with a PDF viewer (`open`) and edit markdown notes with your editor (`ed`).
>> ls 'Why Not Both' | open | ed
+------------------------------------------+--------------+-------+------+
|                   title                  | first author | venue | year |
+========================================================================+
| Ship Compute or Ship Data? Why Not Both? | Jie You      | NSDI  | 2021 |
+------------------------------------------+--------------+-------+------+
>> # Format your markdown notes into an HTML book, and open it in your browser.
>> ls 'Deep Learning' | printf
+------------------------------------------------------------+--------------+---------+------+
|                            title                           | first author |  venue  | year |
+============================================================================================+
| Tiresias: A GPU Cluster Manager for Distributed Deep       | Juncheng Gu  | NSDI    | 2019 |
| Learning                                                   |              |         |      |
|------------------------------------------------------------+--------------+---------+------|
| Nimble: Lightweight and Parallel GPU Task Scheduling for   | Woosuk Kwon  | NeurIPS | 2020 |
| Deep Learning                                              |              |         |      |
+------------------------------------------------------------+--------------+---------+------+

printf

printf will create an HTML book from your notes.

Commands

Invoking reason will start a new command prompt. It accepts unix-like commands that instead work on research papers in your paperbase.

Works now:

• ls filters and prints papers in table format. Default columns are title, first author(by1), venue(at), and year(in).
• cd adds an AND filter to the default set of filters (which is empty upon startup).
• pwd shows the current default filter set by cd.
• touch creates a new entry in your paperbase.
• curl imports papers from the web, e.g. arXiv or usenix.org. It also downloads paper PDFs if available. Also experimentally supports downloading raw PDF urls and inferring metadata fields.
• rm removes entries from your paperbase.
• set sets paper attributes, including custom labels that can also be used to give colors to papers in ls.
• printf creates an HTML page of your notes using mdbook.
• open opens the paper with your PDF viewer (configurable, defaults to zathura).
• ed opens your editor (configurable, defaults to vim), in which you can edit your notes.
• wc counts the number of papers.
• man plus a command will print documentation for that command.
• exit or Ctrl-d quits reason.

Not yet, but hopefully soon (Contributions are more than welcome!):

• grep returns a list of papers whose notes contain the query string that you specify.
• sort sorts papers by given columns.
• stat prints the metadata and notes of papers.
• top prints out a summary of your paperbase.

Installation

You can grab binaries from Releases, or you can run cargo install reason-shell.

Cross-Platform Support

reason currently supports Linux and MacOS. Windows is excluded because the owner does not currently own a Windows machine.

In order to share data between multiple platforms, users are encouraged to locate reason metadata, PDF files, and markdown notes in a location synced by cloud storage services such as Google Drive. I use the official Google Drive app on MacOS and Insync on Linux. This offers an extra benefit - you can read PDFs with your iPad, also synced with the cloud storage.

Documentation

If you already have reason, run man man to view the top-level documentation.

If you're just exploring whether to use reason, take a look at the man directory.

Configuration

The configuration file is kept at ~/.config/reason/config.toml. If not present, reason will generate one populated with default settings.

For more information, open reason and run man config, or read man/config.md.