常见的命令行参数

• --max-arity=2 或 -a2 用于控制找到的抽象的最大算子数量（默认为2）
• --iterations=10 或 -i10 用于控制压缩运行迭代次数。每次迭代产生一个抽象（可以基于前一个抽象）
• --threads=10 或 -t10 是通过多线程提高性能的快捷方式（默认为1）

所有命令行参数

从 cargo run --release --bin=compress -- --help

ARGS:
    <FILE>    json file to read compression input programs from

OPTIONS:
    -a, --max-arity <MAX_ARITY>
            max arity of abstractions to find (will find all from 0 to this number inclusive)
            [default: 2]

        --args-from-json
            extracts argument values from the json; specifically assumes a key value pair like
            "stitch_args": "data/dc/logo_iteration_1_stitchargs.json -a3 -t8 --fmt=dreamcoder
            --dreamcoder-drop-last --no-mismatch-check", in the toplevel dictionary of the json. All
            other commandline args get discarded when you specify this option

    -b, --batch <BATCH>
            how many worklist items a thread will take at once [default: 1]

        --dreamcoder-comparison
            anything related to running a dreamcoder comparison

        --dynamic-batch
            threads will autoadjust how large their batches are based on the worklist size

        --fmt <FMT>
            the format of the input file, e.g. 'programs-list' for a simple JSON array of programs
            or 'dreamcoder' for a JSON in the style expected by the original dreamcoder codebase.
            See [formats.rs] for options or to add new ones [default: programs-list] [possible
            values: dreamcoder, programs-list, split-programs-list]

        --follow-track
            for debugging: prunes all branches except the one that leads to the `--track`
            abstraction

    -h, --help
            Print help information

        --hole-choice <HOLE_CHOICE>
            Method for choosing hole to expand at each step, doesn't have a huge effect [default:
            depth-first] [possible values: random, breadth-first, depth-first, max-largest-subset,
            high-entropy, low-entropy, max-cost, min-cost, many-groups, few-groups, few-apps]

    -i, --iterations <ITERATIONS>
            Number of iterations to run compression for (number of inventions to find) [default: 3]

    -n, --inv-candidates <INV_CANDIDATES>
            Number of invention candidates compression_step should return in a *single* step. Note
            that these will be the top n optimal candidates modulo subsumption pruning (and the top-
            1  is guaranteed to be globally optimal) [default: 1]

        --no-mismatch-check
            disables the safety check for the utility being correct; you only want to do this if you
            truly dont mind unsoundness for a minute

        --no-opt
            disable all optimizations

        --no-opt-arity-zero
            disable the arity zero priming optimization

        --no-opt-force-multiuse
            disable the force multiuse pruning optimization

        --no-opt-free-vars
            disable the free variable pruning optimization

        --no-opt-single-task
            disable the single task pruning optimization

        --no-opt-single-use
            disable the single structurally hashed subtree match pruning

        --no-opt-upper-bound
            disable the upper bound pruning optimization

        --no-opt-useless-abstract
            disable the useless abstraction pruning optimization

        --no-other-util
            makes it so utility is based purely on corpus size without adding in the abstraction
            size

        --no-stats
            Disable stat logging - note that stat logging in multithreading requires taking a mutex
            so it can be a source of slowdown in the massively multithreaded case, hence this flag
            to disable it

        --no-top-lambda
            makes it so inventions cant start with a lambda at the top

    -o, --out <OUT>
            json output file [default: out/out.json]

        --print-stats <PRINT_STATS>
            print stats this often (0 means never) [default: 0]

    -r, --show-rewritten
            print out programs rewritten under abstraction

        --rewrite-check
            whenever you finish an invention do a full rewrite to check that rewriting doesnt raise
            a cost mismatch exception

        --save-rewritten <SAVE_REWRITTEN>
            saves the rewritten frontiers in an input-readable format

        --shuffle
            shuffle order of set of inventions

    -t, --threads <THREADS>
            number of threads (no parallelism if set to 1) [default: 1]

        --track <TRACK>
            for debugging: pattern or abstraction to track

        --truncate <TRUNCATE>
            truncate set of inventions to include only this many (happens after shuffle if shuffle
            is also specified)

        --utility-by-rewrite
            calculate utility exhaustively by performing a full rewrite; mainly used when cost
            mismatches are happening and we need something slow but accurate

        --verbose-best
            prints whenever a new best abstraction is found

        --verbose-worklist
            prints every worklist item as it is processed (will slow things down a ton due to
            rendering out expressins)

禁用优化

cargorun --release --bin=compress --data/cogsci/nuts-bolts.json --no-opt

或者查看以 --no-opt- 开头的其他命令行参数，以禁用特定的优化

Python 绑定

目前提供初始的 Python 绑定。

详细信息

• --save-baseline=main 保存一个命名的基线（如果存在，则与其过去的版本进行比较，然后覆盖它）
• --load-baseline=feature 表示 不运行任何基准测试，只加载文件，就像它是你刚刚生成的结果一样
• --baseline=master 覆盖我们将比较哪个基准
• --bench=compress_bench 避免了详细的“未识别的选项”错误 这里

-->

火焰图

如果你还没有安装： cargo install flamegraph cargo flamegraph --root --open --deterministic --output=out/flamegraph.svg --bin=compress -- data/cogsci/nuts-bolts.json

致谢

这项工作得到了美国国家科学基金会（NSF）的资助，资助编号为 1918839《通过代码理解世界》http://www.neurosymbolic.org/

这项工作部分得到了国防高级研究计划局（DARPA）的资助，资助项目为 Symbiotic Design for Cyber Physical Systems（SDCPS），合同编号 FA8750-20-C-0542（Systemic Generative Engineering）。所表达的观点、意见和/或发现是作者的观点，不一定反映DARPA的观点。