README.md

metadata

license: bsd-3-clause
configs:
  - config_name: computer_vision
    data_files:
      - split: cifar10
        path: experiments_cifar10.csv
      - split: cifar100
        path: experiments_cifar100.csv
      - split: dtd
        path: experiments_dtd.csv
  - config_name: libsvm
    data_files:
      - split: libsvm
        path: experiments_libsvm.csv
  - config_name: roberta
    data_files:
      - split: bookwiki
        path: experiments_roberta_bookwiki.csv

Revisiting Learning Rate Control

This dataset represents the experimental data collected from the paper Revisiting Learning Rate Control. We provide splits for computer_vision, libsvm and roberta experiments. The details for each split are as follows:

computer_vision

$$\text{epoch, seed, method}\mapsto \text{training\_loss, validation\_loss, validation\_accuracy}\backslash text\{epoch,\; seed,\; method\}\; \backslash mapsto\; \backslash text\{training\backslash \_loss,\; validation\backslash \_loss,\; validation\backslash \_accuracy\}$$

libsvm

$$\text{epoch, seed, method, dataset}\mapsto \text{training\_loss, validation\_loss, validation\_accuracy}\backslash text\{epoch,\; seed,\; method,\; dataset\}\; \backslash mapsto\; \backslash text\{training\backslash \_loss,\; validation\backslash \_loss,\; validation\backslash \_accuracy\}$$

roberta

$$\text{step, seed, method}\mapsto \text{training\_perplexity, validation\_perplexity}\backslash text\{step,\; seed,\; method\}\; \backslash mapsto\; \backslash text\{training\backslash \_perplexity,\; validation\backslash \_perplexity\}$$