Skip to content

Latest commit

ย 

History

History
69 lines (61 loc) ยท 4.74 KB

File metadata and controls

69 lines (61 loc) ยท 4.74 KB

KUDataRepresentation

  • ์›๋ณธ ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ๋ฅผ ์ž…๋ ฅ์œผ๋กœ ๋ฐ›์•„ ๊ฐ ๊ด€์ธก์น˜์— ๋Œ€ํ•œ representation vector๋ฅผ ๋„์ถœํ•˜๋Š” time series representation์— ๋Œ€ํ•œ ์„ค๋ช…
  • ์ž…๋ ฅ ๋ฐ์ดํ„ฐ ํ˜•ํƒœ : (num_of_instance x input_dims x seq_len) ์ฐจ์›์˜ ๋‹ค๋ณ€๋Ÿ‰ ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ(multivariate time-series data)

Time series representation ์‚ฌ์šฉ ์‹œ, ์„ค์ •ํ•ด์•ผํ•˜๋Š” ๊ฐ’

  • model : ['ts2vec', 'ts_tcc', 'rae_mepc', 'stoc'] ์ค‘ ์„ ํƒ

  • training : ๋ชจ๋ธ ํ•™์Šต ์—ฌ๋ถ€, [True, False] ์ค‘ ์„ ํƒ, ํ•™์Šต ์™„๋ฃŒ๋œ ๋ชจ๋ธ์ด ์ €์žฅ๋˜์–ด ์žˆ๋‹ค๋ฉด False ์„ ํƒ

  • best_model_path : ํ•™์Šต ์™„๋ฃŒ๋œ ๋ชจ๋ธ์„ ์ €์žฅํ•  ๊ฒฝ๋กœ

  • ์‹œ๊ณ„์—ด representation ๋ชจ๋ธ hyperparameter : ์•„๋ž˜์— ์ž์„ธํžˆ ์„ค๋ช….

    • TS2Vec hyperparameter
    • TS-TCC hyperparameter
    • RAE-MEPC hyperparameter
    • STOC hyperparameter

์‹œ๊ณ„์—ด representation ๋ชจ๋ธ hyperparameter

1. TS2Vec

  • input_dim : ๋ฐ์ดํ„ฐ์˜ ๋ณ€์ˆ˜ ๊ฐœ์ˆ˜, int
  • repr_dim : data representation ์ฐจ์›, int(default: 64, ๋ฒ”์œ„: 1 ์ด์ƒ, 2์˜ ์ง€์ˆ˜๋กœ ์„ค์ • ๊ถŒ์žฅ)
  • hidden_dim : encoder์˜ hidden dimension, int(default: 64, ๋ฒ”์œ„: 1 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • num_epochs : ํ•™์Šต epoch ํšŸ์ˆ˜, int(default: 50, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • batch_size : batch ํฌ๊ธฐ, int(default: 512, ๋ฒ”์œ„: 1 ์ด์ƒ, ์ปดํ“จํ„ฐ ์‚ฌ์–‘์— ์ ํ•ฉํ•˜๊ฒŒ ์„ค์ •)
  • lr : learning rate, float(default: 0.001, ๋ฒ”์œ„: 0.1 ์ดํ•˜)
  • device : ํ•™์Šต ํ™˜๊ฒฝ, (default: 'cuda', ['cuda', 'cpu'] ์ค‘ ์„ ํƒ)

2. TS-TCC

  • input_dim : ๋ฐ์ดํ„ฐ์˜ ๋ณ€์ˆ˜ ๊ฐœ์ˆ˜, int
  • repr_dim : data representation ์ฐจ์›, int(default: 64, ๋ฒ”์œ„: 1 ์ด์ƒ, 2์˜ ์ง€์ˆ˜๋กœ ์„ค์ • ๊ถŒ์žฅ)
  • hidden_dim : temporal / contextual contrasting ๋ชจ๋“ˆ์˜ hidden dimension, int(default: 100, ๋ฒ”์œ„: 1 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • timesteps : temporal contrasting ๋ชจ๋“ˆ์—์„œ ๋ฏธ๋ž˜ ์˜ˆ์ธกํ•  ์‹œ์ ์˜ ๊ธธ์ด, int(default: 6, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • num_epochs : ํ•™์Šต epoch ํšŸ์ˆ˜, int(default: 50, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • batch_size : batch ํฌ๊ธฐ, int(default: 512, ๋ฒ”์œ„: 1 ์ด์ƒ, ์ปดํ“จํ„ฐ ์‚ฌ์–‘์— ์ ํ•ฉํ•˜๊ฒŒ ์„ค์ •)
  • lr : learning rate, float(default: 0.001, ๋ฒ”์œ„: 0.1 ์ดํ•˜)
  • device : ํ•™์Šต ํ™˜๊ฒฝ, (default: 'cuda', ['cuda', 'cpu'] ์ค‘ ์„ ํƒ)
  • jitter_scale_ratio : time series data augementation ์ค‘ weak augementation์˜ ๊ฐ•๋„, float(default: 1.1, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • jitter_ratio : time series data augementation ์ค‘ strong augementation์˜ ๊ฐ•๋„, float(default: 0.8, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • max_seg : strong augementation์—์„œ permutation ์ง„ํ–‰์‹œ ๋ฐ์ดํ„ฐ์˜ ์ตœ๋Œ€ ๋ถ„ํ•  ๊ฐœ์ˆ˜, int(default: 8, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)

3. RAE-MEPC

  • window_size : ๋ชจ๋ธ์˜ input sequence ๊ธธ์ด, int(default: 32, ๋ฒ”์œ„: 0 ์ด์ƒ & ์›๋ž˜ ๋ฐ์ดํ„ฐ์˜ sequence ๊ธธ์ด ์ดํ•˜)
  • input_dim : ๋ฐ์ดํ„ฐ์˜ ๋ณ€์ˆ˜ ๊ฐœ์ˆ˜, int
  • repr_dim : data representation ์ฐจ์›, int(default: 64, ๋ฒ”์œ„: 1 ์ด์ƒ, 2์˜ ์ง€์ˆ˜๋กœ ์„ค์ • ๊ถŒ์žฅ)
  • enc_nlayers : multi-resolution encoder๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” sub-encoder์˜ ๊ฐœ์ˆ˜, int(default: 3, ๋ฒ”์œ„: 1 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • dec_nlayers : multi-resolution decoder๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” sub-decoder์˜ ๊ฐœ์ˆ˜, int(default: 3, ๋ฒ”์œ„: 1 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • tau : multi-resolution encoder ๋ฐ decoder์˜ resolution๋ฅผ ์กฐ์ ˆํ•˜๋Š” ๊ฐ’, int(default: 4, ๋ฒ”์œ„: 2 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • num_epochs : ํ•™์Šต epoch ํšŸ์ˆ˜, int(default: 50, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • batch_size : batch ํฌ๊ธฐ, int(default: 512, ๋ฒ”์œ„: 1 ์ด์ƒ, ์ปดํ“จํ„ฐ ์‚ฌ์–‘์— ์ ํ•ฉํ•˜๊ฒŒ ์„ค์ •)
  • lr : learning rate, float(default: 0.001, ๋ฒ”์œ„: 0.1 ์ดํ•˜)
  • device : ํ•™์Šต ํ™˜๊ฒฝ, (default: 'cuda', ['cuda', 'cpu'] ์ค‘ ์„ ํƒ)

4. STOC

  • window_size : ๋ชจ๋ธ์˜ input sequence ๊ธธ์ด, int(default: 32, ๋ฒ”์œ„: 0 ์ด์ƒ & ์›๋ž˜ ๋ฐ์ดํ„ฐ์˜ sequence ๊ธธ์ด ์ดํ•˜)
  • input_dim : ๋ฐ์ดํ„ฐ์˜ ๋ณ€์ˆ˜ ๊ฐœ์ˆ˜, int
  • repr_dim : data representation ์ฐจ์›, int(default: 64, ๋ฒ”์œ„: 1 ์ด์ƒ, 2์˜ ์ง€์ˆ˜๋กœ ์„ค์ • ๊ถŒ์žฅ)
  • hidden_dim : encoder์˜ hidden dimension, int(default: 256, ๋ฒ”์œ„: 1 ์ด์ƒ, default ๊ฐ’ ์‚ฌ์šฉ ๊ถŒ์žฅ)
  • forecast_step : ๋ฏธ๋ž˜ ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•˜์—ฌ ์˜ˆ์ธกํ•  ์‹œ์ ์˜ ๊ธธ์ด, int(default: 6, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • num_epochs : ํ•™์Šต epoch ํšŸ์ˆ˜, int(default: 50, ๋ฒ”์œ„: 1 ์ด์ƒ)
  • batch_size : batch ํฌ๊ธฐ, int(default: 512, ๋ฒ”์œ„: 1 ์ด์ƒ, ์ปดํ“จํ„ฐ ์‚ฌ์–‘์— ์ ํ•ฉํ•˜๊ฒŒ ์„ค์ •)
  • lr : learning rate, float(default: 0.001, ๋ฒ”์œ„: 0.1 ์ดํ•˜)
  • device : ํ•™์Šต ํ™˜๊ฒฝ, (default: 'cuda', ['cuda', 'cpu'] ์ค‘ ์„ ํƒ)
  • patience : ์˜ˆ์ธก ๋ชจ๋ธ ํ•™์Šต ์‹œ, ์‚ฌ์ „ ์„ค์ •ํ•œ epoch ๋™์•ˆ loss๊ฐ€ ๊ฐ์†Œํ•˜์ง€ ์•Š์œผ๋ฉด ํ•™์Šต ์กฐ๊ธฐ ์ค‘๋‹จ, int(default: 10, ๋ฒ”์œ„: 1 ์ด์ƒ num_epochs ๋ฏธ๋งŒ)