Datasets:
File size: 9,867 Bytes
84a242a | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 | # Copyright 2021 Zhongyang Zhang
# Contact: mirakuruyoo@gmai.com
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" This main entrance of the whole project.
Most of the code should not be changed, please directly
add all the input arguments of your model's constructor
and the dataset file's constructor. The MInterface and
DInterface can be seen as transparent to all your args.
"""
import torch
import pandas as pd
import numpy as np
from argparse import ArgumentParser
torch.backends.cudnn.benchmark = True # Enable for better performance
torch.backends.cudnn.deterministic = False # Allow for better performance
import numpy as np
from typing import Dict
from statsmodels.stats.contingency_tables import mcnemar
import os
from tabulate import tabulate
def compare_binary_selectors(
yA: np.ndarray, yB: np.ndarray, n_boot: int = 10000, seed: int = 0
) -> Dict[str, float]:
"""
Compare two binary docking success outputs (per-instance) with paired statistics.
Args:
yA, yB : arrays of shape (n,) with {0,1} entries (binary success indicators).
n_boot : number of bootstrap resamples for CI.
seed : random seed for reproducibility.
Returns:
dict with:
- mean_A: mean success of A
- mean_B: mean success of B
- mean_diff: mean(A-B)
- ci_low, ci_high: bootstrap 95% CI of mean_diff
- p_mcnemar: exact McNemar test p-value (two-sided)
- n: number of paired instances
"""
yA = np.asarray(yA).astype(float)
yB = np.asarray(yB).astype(float)
assert yA.shape == yB.shape
n = len(yA)
diff = yA - yB
mean_A = float(yA.mean())
mean_B = float(yB.mean())
mean_diff = float(diff.mean())
# Bootstrap CI on mean difference
rng = np.random.default_rng(seed)
boot_means = [diff[rng.integers(0, n, n)].mean() for _ in range(n_boot)]
ci_low, ci_high = np.percentile(boot_means, [2.5, 97.5])
# McNemar’s test
n01 = int(((yA == 0) & (yB == 1)).sum())
n10 = int(((yA == 1) & (yB == 0)).sum())
print(n01)
print(n10)
table = [[0, n01], [n10, 0]]
result = mcnemar(table, exact=True)
pval = result.pvalue
return {
"n": n,
"mean_A": mean_A,
"mean_B": mean_B,
"mean_diff": mean_diff,
"ci_low": ci_low,
"ci_high": ci_high,
"p_mcnemar": pval,
}
def compare_continuous_selectors(
yA: np.ndarray, yB: np.ndarray, n_boot: int = 10000, seed: int = 0
) -> Dict[str, float]:
"""
Compare two continuous docking quality outputs (per-instance) with paired statistics.
Args:
yA, yB : arrays of shape (n,) with continuous entries (e.g., RMSD).
n_boot : number of bootstrap resamples for CI.
seed : random seed for reproducibility.
Returns:
dict with:
- mean_A: mean quality of A
- mean_B: mean quality of B
- mean_diff: mean(A-B)
- ci_low, ci_high: bootstrap 95% CI of mean_diff
- p_mcnemar: exact McNemar test p-value (two-sided)
"""
yA = np.asarray(yA).astype(float)
yB = np.asarray(yB).astype(float)
assert yA.shape == yB.shape
n = len(yA)
diff = yA - yB
mean_A = float(yA.mean())
mean_B = float(yB.mean())
mean_diff = float(diff.mean())
# Bootstrap CI on mean difference
rng = np.random.default_rng(seed)
boot_means = [diff[rng.integers(0, n, n)].mean() for _ in range(n_boot)]
ci_low, ci_high = np.percentile(boot_means, [2.5, 97.5])
# McNemar’s test
n01 = int(((yA < 0.50124) & (yB > 0.50124)).sum())
n10 = int(((yA > 0.50124) & (yB < 0.50124)).sum())
print(n01)
print(n10)
table = [[0, n01], [n10, 0]]
result = mcnemar(table, exact=True)
pval = result.pvalue
return {
"n": n,
"mean_A": mean_A,
"mean_B": mean_B,
"mean_diff": mean_diff,
"ci_low": ci_low,
"ci_high": ci_high,
"p_mcnemar": pval,
}
def main(args):
assert len(args.after_ckpts) > 0, "Please provide at least one --after_ckpts checkpoint."
if len(args.original_ckpts) > 0:
assert len(args.original_ckpts) == len(args.after_ckpts), "Please provide the same number of original and after finetune checkpoints."
#--------------------Add-on: averaged results over multiple seeds---------------------#
eval_results = [pd.read_csv(ckpt[:-5]+'_full.csv') for ckpt in args.after_ckpts]
eval_results = pd.concat(eval_results).groupby('Unnamed: 0').mean()
print("Averaged results over multiple seeds:")
# Recompute the column VBS_SBS_Gap_Closed_by_AS(%)
# How to extract the second column without using its name? Answer: use iloc
sbs_col_name = eval_results.columns[-4]
eval_results['VBS_SBS_Gap_Closed_by_AS(%)'] = (eval_results['AS'] - eval_results[sbs_col_name]) / (eval_results['oracle'] - eval_results[sbs_col_name]) * 100.0
# Print proper table
# print(tabulate(eval_results, headers='keys', tablefmt='psql'))
# Print the transposed version for better readability
print("\nTransposed view:")
print(tabulate(eval_results.T, headers='keys', tablefmt='psql'))
if args.summary_csv is not None:
os.makedirs(os.path.dirname(args.summary_csv), exist_ok=True)
eval_results.T.to_csv(args.summary_csv)
#------------------------------------------------------------------------------------#
algorithm_counts = [pd.read_csv(ckpt[:-5]+'_algorithm_selection_counts.csv') for ckpt in args.after_ckpts]
algorithm_counts = pd.concat(algorithm_counts).groupby('Algorithm').sum()
# Divide by column total to get percentages
algorithm_counts = algorithm_counts.div(algorithm_counts.sum(axis=0), axis=1) * 100.0
descriptives = [pd.read_csv(ckpt[:-5]+'_descriptive.csv') for ckpt in args.after_ckpts]
# Merge descriptives
df_full = pd.concat(descriptives).reset_index(drop=True)
# Drop first column if it's unnamed or not
if df_full.columns[0] == 'Unnamed: 0' or not df_full.columns[0]:
df_full = df_full.drop(df_full.columns[0], axis=1)
dict_as_vs_sbs = compare_continuous_selectors(df_full[sbs_col_name], df_full["AS"])
print(df_full[sbs_col_name], df_full["AS"])
print("Comparison of Algorithm Selection vs SBS:")
print(dict_as_vs_sbs)
def precision_when_selected(df):
oracle = df["oracle_choose"]
choose = df["choose"]
precision = []
for alg in algorithm_counts.index:
picks = (choose == alg)
if picks.sum() == 0:
precision.append(np.nan)
else:
precision.append((oracle[picks] == choose[picks]).mean()*100)
return precision
algorithm_counts['Precision_when_selected(%)'] = precision_when_selected(df_full)
algorithm_counts['Correctly_chosen(%)'] = (algorithm_counts['Precision_when_selected(%)'] * algorithm_counts['Selection_Count']) / 100.0
print("\nAveraged Algorithm Selection Counts over ckpts:")
print(tabulate(algorithm_counts, headers='keys', tablefmt='psql'))
if args.algorithm_counts_csv is not None:
os.makedirs(os.path.dirname(args.algorithm_counts_csv), exist_ok=True)
algorithm_counts.to_csv(args.algorithm_counts_csv)
#------------------------------------------------------------------------------------#
if len(args.original_ckpts) > 0:
original_2_results = [np.load(ckpt[:-5]+'_2_threshold.npy') for ckpt in args.original_ckpts]
original_1_results = [np.load(ckpt[:-5]+'_1_threshold.npy') for ckpt in args.original_ckpts]
after_2_results = [np.load(ckpt[:-5]+'_2_threshold.npy') for ckpt in args.after_ckpts]
after_1_results = [np.load(ckpt[:-5]+'_1_threshold.npy') for ckpt in args.after_ckpts]
print(f'Shapes of original npys: {[arr.shape for arr in original_2_results]}, {[arr.shape for arr in original_1_results]}')
# Average over seeds
original_2_results = np.concatenate(original_2_results).flatten()
original_1_results = np.concatenate(original_1_results).flatten()
after_2_results = np.concatenate(after_2_results).flatten()
after_1_results = np.concatenate(after_1_results).flatten()
dict_2 = compare_binary_selectors(original_2_results, after_2_results)
dict_1 = compare_binary_selectors(original_1_results, after_1_results)
print("Comparison of original vs after finetune (RMSD < 2A & valid pose):")
print(dict_2)
print("Comparison of original vs after finetune (RMSD < 1A & valid pose):")
print(dict_1)
else:
print("Skipped original-vs-after comparison because --original_ckpts was not provided.")
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('--original_ckpts', default=[], type=str, nargs='*', help='Original checkpoints')
parser.add_argument('--after_ckpts', default=[], type=str, nargs='*', help='After finetune checkpoints')
parser.add_argument('--summary_csv', default=None, type=str, help='Optional output CSV for transposed summary table')
parser.add_argument('--algorithm_counts_csv', default=None, type=str, help='Optional output CSV for averaged algorithm count table')
args = parser.parse_args()
main(args) |