File size: 3,196 Bytes
1bb8d13 |
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 |
import os
import matplotlib as mpl
import matplotlib.pyplot as plt
import pandas as pd
from random import shuffle, seed
# LM
model_size = {
"T5\textsubscript{SMALL}": [60, "T5"],
"T5\textsubscript{BASE}": [200, "T5"],
"T5\textsubscript{LARGE}": [770, "T5"],
"T5\textsubscript{XL}": [3000, "T5"],
"T5\textsubscript{XXL}": [11000, "T5"],
"Flan-T5\textsubscript{SMALL}": [60, "Flan-T5"],
"Flan-T5\textsubscript{BASE}": [200, "Flan-T5"],
"Flan-T5\textsubscript{LARGE}": [770, "Flan-T5"],
"Flan-T5\textsubscript{XL}": [3000, "Flan-T5"],
"Flan-T5\textsubscript{XXL}": [11000, "Flan-T5"],
"OPT\textsubscript{125M}": [125, "OPT"],
"OPT\textsubscript{350M}": [350, "OPT"],
"OPT\textsubscript{1.3B}": [1300, "OPT"],
"OPT\textsubscript{2.7B}": [2700, "OPT"],
"OPT\textsubscript{6.7B}": [6700, "OPT"],
"OPT\textsubscript{13B}": [13000, "OPT"],
"OPT\textsubscript{30B}": [30000, "OPT"],
"OPT\textsubscript{66B}": [66000, "OPT"],
"OPT-IML\textsubscript{1.3B}": [1300, "OPT-IML"],
"OPT-IML\textsubscript{30B}": [30000, "OPT-IML"],
}
lm_list = ['T5', 'Flan-T5', 'OPT', 'OPT-IML']
# Oracle
df_oracle = pd.read_csv("experiments/results/oracle.csv", index_col=0)
# Plot
os.makedirs('experiments/figures/main', exist_ok=True)
plt.rcParams.update({'font.size': 18}) # must set in top
def main(target_relation: str = "average", prompt_type: str = "lc"):
df = pd.read_csv(f"experiments/results/lm_{prompt_type}/lm.csv")
df.index = df.pop("model")
df = (df * 100).round(1)
df = df[[i in model_size for i in df.index]]
df["size"] = [model_size[i][0] * 1000000 for i in df.index]
df["lm"] = [model_size[i][1] for i in df.index]
df_target = df[[target_relation, "size", "lm"]]
out = df_target.pivot_table(index='size', columns='lm', aggfunc='mean')
out.columns = [i[1] for i in out.columns]
out = out.reset_index()
out = out[['size'] + lm_list]
styles = ['o-', '^--', 'X:', "P:"]
seed(1)
colors = list(mpl.colormaps['tab20b'].colors)
shuffle(colors)
ax = None
for n, c in enumerate(lm_list):
tmp = out[['size', c]].dropna().reset_index()
ax = tmp.plot.line(y=c,
x='size',
xlabel='',
ylabel="",
ax=ax,
color=colors[n],
style=styles[n],
label=c,
logx=True,
grid=True,
figsize=(4, 5))
if prompt_type == 'qa':
ax.legend().remove()
else:
ax.legend(loc='lower right')
# ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))
plt.tight_layout()
plt.savefig(f"experiments/figures/main/{prompt_type}.{target_relation.replace(' ', '_').replace('/', '-')}.png", bbox_inches="tight", dpi=600)
if __name__ == '__main__':
for p in ['lc', 'qa']:
main('average', p)
main("is competitor/rival of", p)
main("is friend/ally of", p)
main("is influenced by", p)
main("is known for", p)
main("is similar to", p)
|