tap_layer_scaling.json

{
 "question": "does the L20 advantage over L47 survive more training data",
 "holdout": 1000,
 "seeds_per_point": 5,
 "results": {
  "L20": {
   "250": {
    "mean_r@1": 0.06620000451803207,
    "std": 0.009495263234598202,
    "seeds": [
     0.07300000637769699,
     0.0690000057220459,
     0.049000002443790436,
     0.06400000303983688,
     0.07600000500679016
    ]
   },
   "500": {
    "mean_r@1": 0.10200000554323196,
    "std": 0.009055386540152885,
    "seeds": [
     0.0860000029206276,
     0.11000000685453415,
     0.09800000488758087,
     0.10900000482797623,
     0.10700000822544098
    ]
   },
   "1000": {
    "mean_r@1": 0.1674000084400177,
    "std": 0.0061514241166914455,
    "seeds": [
     0.16600000858306885,
     0.15700000524520874,
     0.17500001192092896,
     0.16700001060962677,
     0.1720000058412552
    ]
   },
   "2000": {
    "mean_r@1": 0.2190000146627426,
    "std": 0.011644741355641638,
    "seeds": [
     0.22200001776218414,
     0.1990000158548355,
     0.23500001430511475,
     0.22200001776218414,
     0.21700000762939453
    ]
   },
   "4000": {
    "mean_r@1": 0.28880001306533815,
    "std": 0.007934730169064928,
    "seeds": [
     0.30000001192092896,
     0.28600001335144043,
     0.29600000381469727,
     0.27900001406669617,
     0.28300002217292786
    ]
   }
  },
  "L47": {
   "250": {
    "mean_r@1": 0.056800002604722975,
    "std": 0.0059799675602018616,
    "seeds": [
     0.06200000271201134,
     0.049000002443790436,
     0.0560000017285347,
     0.0650000050663948,
     0.052000001072883606
    ]
   },
   "500": {
    "mean_r@1": 0.08860000371932983,
    "std": 0.006216108670602526,
    "seeds": [
     0.09400000423192978,
     0.08300000429153442,
     0.09800000488758087,
     0.08300000429153442,
     0.08500000089406967
    ]
   },
   "1000": {
    "mean_r@1": 0.13700000643730165,
    "std": 0.004049695137703997,
    "seeds": [
     0.13900001347064972,
     0.1380000114440918,
     0.14300000667572021,
     0.13300000131130219,
     0.13199999928474426
    ]
   },
   "2000": {
    "mean_r@1": 0.1868000090122223,
    "std": 0.0053065998604589475,
    "seeds": [
     0.18000000715255737,
     0.18400001525878906,
     0.19600000977516174,
     0.1860000044107437,
     0.18800000846385956
    ]
   },
   "4000": {
    "mean_r@1": 0.2282000094652176,
    "std": 0.00331058983915294,
    "seeds": [
     0.22300000488758087,
     0.22800001502037048,
     0.23000000417232513,
     0.2330000102519989,
     0.22700001299381256
    ]
   }
  }
 },
 "gap_L20_minus_L47": {
  "250": 0.0094,
  "500": 0.0134,
  "1000": 0.0304,
  "2000": 0.0322,
  "4000": 0.0606
 },
 "gap_note": "mean r@1 at L20 minus mean r@1 at L47 at each training size, 5 seeds, same 1000 held out. Direction replicates on train2017 (tap_layer_scaling_30k.json) but magnitude does not."
}