File size: 5,087 Bytes
cf5659c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 |
import json
import os
from argparse import ArgumentParser
import numpy as np
from matplotlib import pyplot as plt
def get_args():
parser = ArgumentParser()
parser.add_argument('--input-files', type=lambda s: s.split(','), required=True, help='Input file that hold all evaluation metrics')
return parser.parse_args()
# TODO: fill it up
RANDOM_BASELINE={
"arc_challenge_acc": 0.2502, # Source: https://arxiv.org/pdf/1803.05457.pdf table 6
"arc_easy_acc": 0.2502, # Source: https://arxiv.org/pdf/1803.05457.pdf table 6
"boolq_acc": 0.5,
"copa_acc": 0.5,
"headqa_acc": 0.25, # TODO: That's a pain as some have 4, some have 5 and nobody reports random baseline
"hellaswag_acc": 0.25,
"lambada_acc": 0., # Safe to say that random models won't perform well at all.
"logiqa_acc": 0.25,
"mathqa_acc": 0.25, # TODO: That's a pain as some have 4, some have 5 and nobody reports random baseline
"mrpc_acc": 0.5,
"multirc_acc": 0., # TODO: I couldn't figure it out
"openbookqa_acc": 0.25,
"piqa_acc": 0.5,
"prost_acc": 0.25,
"pubmedqa_acc": 1/3,
"qnli_acc": 0.5,
"qqp_acc": 0.5,
"race_acc": 0.25, # Source: https://arxiv.org/pdf/1704.04683.pdf table 5
"rte_acc": 0.5,
"sciq_acc": 0.25,
"sst_acc": 0.5,
"triviaqa_acc": 0.,
"webqs_acc": 0.,
"wic_acc": 0.5,
"winogrande_acc": 0.5,
"wnli_acc": 0.5,
"wsc_acc": 0.5
}
def normalise_scores(scores_per_task):
normalised_scores = {}
for key,value in scores_per_task.items():
# We assume it exists, otherwise we need to figure out what the random baseline is
normalised_scores[key] = (value - RANDOM_BASELINE[key]) / (1. - RANDOM_BASELINE[key])
# TODO: we need to substract the random baseline.
return scores_per_task
def main():
args = get_args()
final = {}
for input_file in args.input_files:
assert os.path.basename(input_file).endswith("_agg.json")
experiment_name = os.path.basename(input_file).split("_agg.json")[0]
with open(input_file, "r") as fi:
final[experiment_name] = json.load(fi)
# We search for matching tokens
matching_tokens = set(next(iter(final.values()))["tokens"])
for experiment_name, experiment in final.items():
tokens = experiment["tokens"]
matching_tokens = matching_tokens & set(tokens)
# Make sure we don't override existing data
assert "token2checkpoint_step" not in experiment
experiment["token2checkpoint_step"] = {token: ckpt_step for token, ckpt_step in zip(tokens, experiment["checkpoints"])}
# Make sure we don't override existing data
assert "token2id" not in experiment
experiment["token2id"] = {token: _id for _id, token in enumerate(tokens)}
matching_tokens = sorted(matching_tokens)
print(f"Plotting only for tokens in {matching_tokens}")
plots_per_keys = {}
for token in matching_tokens:
for experiment_name, experiment in final.items():
_id = experiment["token2id"][token]
scores_per_task = {
"Average_acc": {
f"{evaluation_name}_{metric_name}": metric[_id]
for evaluation_name, evaluation in experiment["results"].items()
for metric_name, metric in evaluation.items()
if metric_name == "acc"
},
# "Average": {
# metric_name: values[i]
# for evaluation_name in final["results"][experiment_name]
# for metric_name, values in final["results"][experiment_name][evaluation_name].items()
# if metric_name[-7:] != "_stderr"
# }
}
# Build plot graphs
for key in scores_per_task:
if key not in plots_per_keys:
plots_per_keys[key] = {}
plot_per_token = plots_per_keys[key]
if token in plot_per_token:
continue
plot = plt.figure()
plot = plot.add_subplot(1, 1, 1)
plot.set_title(f"{key} - Number of tokens seen: {token}")
plot_per_token[token] = plot
# Plot per steps
for key in plots_per_keys:
scores = scores_per_task[key]
plot = plots_per_keys[key][token]
# Normalize score
normalised_scores = normalise_scores(scores)
# Sort scores, we order them from smalles to biggest
sorted_scores = sorted(normalised_scores.values())
# Compute the number of task over that sorted_scores.
y = np.arange(len(sorted_scores), 0, -1) / len(sorted_scores)
plot.step(x=sorted_scores, y=y, label=experiment_name)
for plots in plots_per_keys.values():
assert len(plots) == len(matching_tokens)
for plot in plots.values():
plot.legend()
plt.show()
if __name__ == "__main__":
main()
|