genrl-enhancer-diffusion / GENERator /src /tasks /downstream /plot_homework_results.py
Ouzhang's picture
Upload folder using huggingface_hub
a80f3c0 verified
Raw
History Blame Contribute Delete
12.2 kB
import argparse
import json
from pathlib import Path
from typing import Any, Dict, List, Optional
import matplotlib.pyplot as plt
DEFAULT_BUCKET_ORDER = ["low", "mid", "high"]
def parse_args():
parser = argparse.ArgumentParser(
description="Plot homework results for the DeepSTARR / GENERator assignments"
)
parser.add_argument(
"--task1_test_results",
type=str,
default="results/deepstarr_regression/test_results.json",
help="Path to task 1 test_results.json",
)
parser.add_argument(
"--task2_generation_summary",
type=str,
default="results/deepstarr_sft_valid/generation_summary.json",
help="Path to task 2 generation_summary.json",
)
parser.add_argument(
"--task3_generation_summary",
type=str,
default="results/deepstarr_sft_conditioned_valid/generation_summary.json",
help="Path to task 3 generation_summary.json",
)
parser.add_argument(
"--task3_scoring_summary",
type=str,
default="results/deepstarr_conditioned_scoring/scoring_summary.json",
help="Path to task 3 scoring_summary.json",
)
parser.add_argument(
"--output_dir",
type=str,
default="report/figures",
help="Directory to save plots",
)
return parser.parse_args()
def load_json(path: str) -> Optional[Dict[str, Any]]:
p = Path(path)
if not p.exists():
print(f"[WARN] Missing file: {p}")
return None
with open(p, "r", encoding="utf-8") as f:
return json.load(f)
def configure_matplotlib():
plt.style.use("seaborn-v0_8-whitegrid")
plt.rcParams["figure.dpi"] = 180
plt.rcParams["savefig.dpi"] = 220
plt.rcParams["font.size"] = 11
plt.rcParams["axes.titlesize"] = 13
plt.rcParams["axes.labelsize"] = 11
def plot_task1_regression_metrics(task1: Dict[str, Any], output_dir: Path) -> Optional[Path]:
pearson_values = [
task1.get("test_pearson_label_0"),
task1.get("test_pearson_label_1"),
task1.get("test_pearson"),
]
r2_values = [
task1.get("test_r2_label_0"),
task1.get("test_r2_label_1"),
task1.get("test_r2"),
]
if any(value is None for value in pearson_values + r2_values):
return None
labels = ["Label 0", "Label 1", "Overall"]
colors = ["#2C7FB8", "#41B6C4", "#253494"]
fig, axes = plt.subplots(1, 2, figsize=(11.5, 4.6))
for ax, values, title, ylabel in [
(axes[0], pearson_values, "Task 1 Pearson Correlation", "Pearson"),
(axes[1], r2_values, "Task 1 R2 Score", "R2"),
]:
bars = ax.bar(labels, values, color=colors, width=0.62)
ax.set_ylim(0.0, max(0.85, max(values) + 0.08))
ax.set_ylabel(ylabel)
ax.set_title(title)
for bar, value in zip(bars, values):
ax.text(
bar.get_x() + bar.get_width() / 2,
bar.get_height() + 0.015,
f"{value:.3f}",
ha="center",
va="bottom",
)
fig.suptitle("Homework 1: Enhancer Activity Prediction", y=1.02, fontsize=14)
fig.tight_layout()
output_path = output_dir / "task1_regression_metrics.png"
fig.savefig(output_path, bbox_inches="tight")
plt.close(fig)
return output_path
def plot_generation_comparison(
task2: Dict[str, Any], task3: Dict[str, Any], output_dir: Path
) -> Optional[Path]:
metrics = [
("mean_bp_accuracy", "BP Accuracy"),
("valid_dna_rate", "Valid DNA Rate"),
("unique_rate", "Unique Rate"),
]
if any(task2.get(key) is None or task3.get(key) is None for key, _ in metrics):
return None
labels = [label for _, label in metrics]
task2_values = [task2[key] for key, _ in metrics]
task3_values = [task3[key] for key, _ in metrics]
x = list(range(len(labels)))
width = 0.34
fig, ax = plt.subplots(figsize=(8.8, 5.2))
bars1 = ax.bar(
[i - width / 2 for i in x],
task2_values,
width=width,
color="#3182BD",
label="Task 2 Unconditional",
)
bars2 = ax.bar(
[i + width / 2 for i in x],
task3_values,
width=width,
color="#E6550D",
label="Task 3 Conditioned",
)
ax.set_xticks(x)
ax.set_xticklabels(labels)
ax.set_ylim(0.0, 1.1)
ax.set_ylabel("Value")
ax.set_title("Task 2 vs Task 3: Generation Quality Comparison")
ax.legend()
for bars in [bars1, bars2]:
for bar in bars:
ax.text(
bar.get_x() + bar.get_width() / 2,
bar.get_height() + 0.02,
f"{bar.get_height():.3f}",
ha="center",
va="bottom",
fontsize=9,
)
fig.tight_layout()
output_path = output_dir / "task23_generation_comparison.png"
fig.savefig(output_path, bbox_inches="tight")
plt.close(fig)
return output_path
def _get_bucket_summary(task3_scoring: Dict[str, Any]) -> Optional[Dict[str, Dict[str, Any]]]:
if "by_activity_bucket" in task3_scoring:
return task3_scoring["by_activity_bucket"]
if "by_condition_token" in task3_scoring:
return task3_scoring["by_condition_token"]
return None
def _ordered_buckets(bucket_summary: Dict[str, Dict[str, Any]]) -> List[str]:
present = list(bucket_summary.keys())
ordered = [bucket for bucket in DEFAULT_BUCKET_ORDER if bucket in present]
remainder = [bucket for bucket in present if bucket not in ordered]
return ordered + sorted(remainder)
def plot_task3_bucket_prediction_sum(
task3_scoring: Dict[str, Any], output_dir: Path
) -> Optional[Path]:
bucket_summary = _get_bucket_summary(task3_scoring)
if not bucket_summary:
return None
buckets = _ordered_buckets(bucket_summary)
generated = [bucket_summary[b]["mean_generated_prediction_sum"] for b in buckets]
reference = [bucket_summary[b]["mean_reference_prediction_sum"] for b in buckets]
x = list(range(len(buckets)))
width = 0.34
fig, ax = plt.subplots(figsize=(8.6, 5.2))
bars1 = ax.bar(
[i - width / 2 for i in x],
reference,
width=width,
color="#9ECAE1",
label="Reference Sequence",
)
bars2 = ax.bar(
[i + width / 2 for i in x],
generated,
width=width,
color="#08519C",
label="Generated Sequence",
)
ax.set_xticks(x)
ax.set_xticklabels([bucket.capitalize() for bucket in buckets])
ax.set_ylabel("Predictor Score Sum")
ax.set_title("Task 3: Predictor Score by Activity Bucket")
ax.axhline(0.0, color="black", linewidth=0.8)
ax.legend()
for bars in [bars1, bars2]:
for bar in bars:
y = bar.get_height()
offset = 0.03 if y >= 0 else -0.07
ax.text(
bar.get_x() + bar.get_width() / 2,
y + offset,
f"{y:.3f}",
ha="center",
va="bottom" if y >= 0 else "top",
fontsize=9,
)
fig.tight_layout()
output_path = output_dir / "task3_bucket_prediction_sum.png"
fig.savefig(output_path, bbox_inches="tight")
plt.close(fig)
return output_path
def plot_task3_bucket_deltas(
task3_scoring: Dict[str, Any], output_dir: Path
) -> Optional[Path]:
bucket_summary = _get_bucket_summary(task3_scoring)
if not bucket_summary:
return None
buckets = _ordered_buckets(bucket_summary)
delta_l0 = [bucket_summary[b].get("mean_prediction_delta_label_0", 0.0) for b in buckets]
delta_l1 = [bucket_summary[b].get("mean_prediction_delta_label_1", 0.0) for b in buckets]
delta_sum = [bucket_summary[b].get("mean_prediction_delta_sum", 0.0) for b in buckets]
positive_rate = [bucket_summary[b].get("positive_delta_rate", 0.0) for b in buckets]
x = list(range(len(buckets)))
width = 0.25
fig, axes = plt.subplots(1, 2, figsize=(11.2, 4.8))
bars_l0 = axes[0].bar(
[i - width / 2 for i in x],
delta_l0,
width=width,
color="#31A354",
label="Delta Label 0",
)
bars_l1 = axes[0].bar(
[i + width / 2 for i in x],
delta_l1,
width=width,
color="#756BB1",
label="Delta Label 1",
)
axes[0].axhline(0.0, color="black", linewidth=0.8)
axes[0].set_xticks(x)
axes[0].set_xticklabels([bucket.capitalize() for bucket in buckets])
axes[0].set_ylabel("Prediction Delta")
axes[0].set_title("Task 3: Label-wise Delta by Bucket")
axes[0].legend()
for bars in [bars_l0, bars_l1]:
for bar in bars:
y = bar.get_height()
offset = 0.03 if y >= 0 else -0.06
axes[0].text(
bar.get_x() + bar.get_width() / 2,
y + offset,
f"{y:.3f}",
ha="center",
va="bottom" if y >= 0 else "top",
fontsize=8,
)
bars_sum = axes[1].bar(
[i - width / 2 for i in x],
delta_sum,
width=width,
color="#E6550D",
label="Delta Sum",
)
bars_rate = axes[1].bar(
[i + width / 2 for i in x],
positive_rate,
width=width,
color="#FDD0A2",
label="Positive Delta Rate",
)
axes[1].axhline(0.0, color="black", linewidth=0.8)
axes[1].set_xticks(x)
axes[1].set_xticklabels([bucket.capitalize() for bucket in buckets])
axes[1].set_ylabel("Value")
axes[1].set_ylim(min(-0.25, min(delta_sum) - 0.08), max(1.0, max(positive_rate) + 0.1))
axes[1].set_title("Task 3: Overall Shift by Bucket")
axes[1].legend()
for bars in [bars_sum, bars_rate]:
for bar in bars:
y = bar.get_height()
offset = 0.03 if y >= 0 else -0.06
axes[1].text(
bar.get_x() + bar.get_width() / 2,
y + offset,
f"{y:.3f}",
ha="center",
va="bottom" if y >= 0 else "top",
fontsize=8,
)
fig.tight_layout()
output_path = output_dir / "task3_bucket_deltas.png"
fig.savefig(output_path, bbox_inches="tight")
plt.close(fig)
return output_path
def save_compact_snapshot(
task1: Optional[Dict[str, Any]],
task2: Optional[Dict[str, Any]],
task3_gen: Optional[Dict[str, Any]],
task3_score: Optional[Dict[str, Any]],
output_dir: Path,
) -> Path:
snapshot = {
"task1": task1,
"task2_generation": task2,
"task3_generation": task3_gen,
"task3_scoring": task3_score,
}
output_path = output_dir / "metrics_snapshot.json"
with open(output_path, "w", encoding="utf-8") as f:
json.dump(snapshot, f, indent=2, ensure_ascii=False)
return output_path
def main():
args = parse_args()
output_dir = Path(args.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
configure_matplotlib()
task1 = load_json(args.task1_test_results)
task2 = load_json(args.task2_generation_summary)
task3_gen = load_json(args.task3_generation_summary)
task3_score = load_json(args.task3_scoring_summary)
generated_paths = []
if task1 is not None:
path = plot_task1_regression_metrics(task1, output_dir)
if path is not None:
generated_paths.append(path)
if task2 is not None and task3_gen is not None:
path = plot_generation_comparison(task2, task3_gen, output_dir)
if path is not None:
generated_paths.append(path)
if task3_score is not None:
path = plot_task3_bucket_prediction_sum(task3_score, output_dir)
if path is not None:
generated_paths.append(path)
path = plot_task3_bucket_deltas(task3_score, output_dir)
if path is not None:
generated_paths.append(path)
snapshot_path = save_compact_snapshot(task1, task2, task3_gen, task3_score, output_dir)
generated_paths.append(snapshot_path)
print("Generated files:")
for path in generated_paths:
print(f" - {path}")
if __name__ == "__main__":
main()