File size: 12,858 Bytes
2560dd0 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 | """
Ablation analysis β mirrors all_chrom_evaluation.ipynb but for:
- Ablation_no_partition, Ablation_no_priority, Ablation_no_length
- Plus Baseline_bpe_5120 and Merged_uni_len2_5120 as references
Saves all plots to eval_outputs_ablation/plots/
"""
import os
import pandas as pd
import numpy as np
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import seaborn as sns
# ββ Directories ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
ABLATION_DIR = "/home/n5huang/dna_token/tokenizer_evaluation/eval_outputs_ablation"
MERGED_DIR = "/home/n5huang/dna_token/tokenizer_evaluation/eval_outputs_all_chrom"
BASELINE_DIR = "/home/n5huang/dna_token/tokenizer_evaluation/eval_outputs_all_chrom_baseline_bpe"
PLOT_DIR = os.path.join(ABLATION_DIR, "plots")
os.makedirs(PLOT_DIR, exist_ok=True)
pd.set_option("display.max_columns", 20)
pd.set_option("display.width", 200)
pd.set_option("display.float_format", "{:.6f}".format)
# ββ Helper: compute summary row from an agg CSV βββββββββββββββββββββββββββββ
def summarize_agg(agg: pd.DataFrame, name: str, scope: str, region: str = None):
row = {
"tokenizer": name,
"scope": scope,
"mean_of_mean_phyloP": agg["mean_mean"].mean(),
"median_of_mean_phyloP": agg["mean_mean"].median(),
"pct_mean_phyloP_above_0": (agg["mean_mean"] > 0).mean() * 100,
"mean_of_variance": agg["mean_var"].mean(),
"median_of_variance": agg["mean_var"].median(),
"pct_variance_below_0.1": (agg["mean_var"] < 0.1).mean() * 100,
"num_tokens": len(agg),
"mean_token_count": agg["count"].mean(),
"median_token_count": agg["count"].median(),
}
if region:
row["region"] = region
return row
# ββ 1. Window-Based Summary ββββββββββββββββββββββββββββββββββββββββββββββββββ
print("=" * 80)
print("1. WINDOW-BASED SUMMARY")
print("=" * 80)
ablation_names = ["Ablation_no_partition", "Ablation_no_priority", "Ablation_no_length"]
ref_names = {
"Baseline_bpe_5120": os.path.join(BASELINE_DIR, "agg_Baseline_bpe_5120.csv"),
"Merged_uni_len2_5120": os.path.join(MERGED_DIR, "agg_Merged_uni_len2_5120.csv"),
}
# Load all agg data (window)
agg_data = {}
window_summaries = []
for name in ablation_names:
path = os.path.join(ABLATION_DIR, f"agg_{name}.csv")
agg = pd.read_csv(path)
agg_data[name] = agg
window_summaries.append(summarize_agg(agg, name, "all_chrom_window"))
for name, path in ref_names.items():
agg = pd.read_csv(path)
agg_data[name] = agg
window_summaries.append(summarize_agg(agg, name, "all_chrom_window"))
df_window = pd.DataFrame(window_summaries)
cols = ["tokenizer", "mean_of_mean_phyloP", "median_of_mean_phyloP",
"pct_mean_phyloP_above_0", "mean_of_variance", "median_of_variance",
"pct_variance_below_0.1", "num_tokens", "mean_token_count", "median_token_count"]
df_window = df_window[cols]
print(df_window.to_string(index=False))
df_window.to_csv(os.path.join(ABLATION_DIR, "summary_ablation_window.csv"), index=False)
# ββ 2. Region-Based Summary βββββββββββββββββββββββββββββββββββββββββββββββββ
print("\n" + "=" * 80)
print("2. REGION-BASED SUMMARY")
print("=" * 80)
agg_region = {}
region_summaries = []
for name in ablation_names:
for region in ["conserved", "neutral", "accelerated"]:
path = os.path.join(ABLATION_DIR, f"agg_{name}_{region}.csv")
if os.path.exists(path):
agg = pd.read_csv(path)
agg_region[(name, region)] = agg
region_summaries.append(summarize_agg(agg, name, "all_chrom_region", region))
for name, base_path in ref_names.items():
base_dir = os.path.dirname(base_path)
for region in ["conserved", "neutral", "accelerated"]:
path = os.path.join(base_dir, f"agg_{name}_{region}.csv")
if os.path.exists(path):
agg = pd.read_csv(path)
agg_region[(name, region)] = agg
region_summaries.append(summarize_agg(agg, name, "all_chrom_region", region))
df_region = pd.DataFrame(region_summaries)
cols_region = ["tokenizer", "region", "mean_of_mean_phyloP", "median_of_mean_phyloP",
"pct_mean_phyloP_above_0", "mean_of_variance", "median_of_variance",
"pct_variance_below_0.1", "num_tokens", "mean_token_count", "median_token_count"]
df_region = df_region[cols_region]
for region in ["conserved", "neutral", "accelerated"]:
print(f"\n=== {region.upper()} ===")
print(df_region[df_region["region"] == region].to_string(index=False))
df_region.to_csv(os.path.join(ABLATION_DIR, "summary_ablation_all_regions.csv"), index=False)
# ββ Color scheme βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
COLORS = {
"Baseline_bpe_5120": "steelblue",
"Merged_uni_len2_5120": "tomato",
"Ablation_no_partition": "#2ca02c", # green
"Ablation_no_priority": "#9467bd", # purple
"Ablation_no_length": "#ff7f0e", # orange
}
# ββ 3. Scatter: Mean phyloP vs Variance (Window) ββββββββββββββββββββββββββββ
print("\n" + "=" * 80)
print("3. SCATTER PLOTS: Mean phyloP vs Variance")
print("=" * 80)
fig, axes = plt.subplots(1, 2, figsize=(18, 8))
for ax_idx, (xcol, xlabel) in enumerate([("mean_mean", "Mean phyloP"), ("median_mean", "Median Mean phyloP")]):
ax = axes[ax_idx]
for name, color in COLORS.items():
if name in agg_data:
d = agg_data[name]
ax.scatter(d[xcol], d["mean_var"], alpha=0.4, s=20, label=name, color=color)
ax.set_xlabel(xlabel)
ax.set_ylabel("Mean Variance")
ax.set_title(f"Token Conservation: {xlabel} vs Variance")
ax.legend(fontsize=8)
plt.tight_layout()
plt.savefig(os.path.join(PLOT_DIR, "scatter_mean_vs_var.png"), dpi=150, bbox_inches="tight")
plt.close()
print(f" Saved scatter_mean_vs_var.png")
# ββ 4. Histograms: Mean phyloP and Variance distributions βββββββββββββββββββ
print("\n" + "=" * 80)
print("4. HISTOGRAMS")
print("=" * 80)
fig, axes = plt.subplots(1, 2, figsize=(18, 6))
ax = axes[0]
for name, color in COLORS.items():
if name in agg_data:
ax.hist(agg_data[name]["mean_mean"], bins=50, alpha=0.5, label=name, color=color)
ax.set_xlabel("Mean phyloP")
ax.set_ylabel("Token Count")
ax.set_title("Mean Conservation Distribution")
ax.legend(fontsize=8)
ax = axes[1]
for name, color in COLORS.items():
if name in agg_data:
ax.hist(agg_data[name]["mean_var"], bins=50, alpha=0.5, label=name, color=color)
ax.set_xlabel("Mean Variance")
ax.set_ylabel("Token Count")
ax.set_title("Variance Distribution")
ax.legend(fontsize=8)
plt.tight_layout()
plt.savefig(os.path.join(PLOT_DIR, "histograms_mean_var.png"), dpi=150, bbox_inches="tight")
plt.close()
print(f" Saved histograms_mean_var.png")
# ββ 5. Region scatter: by region βββββββββββββββββββββββββββββββββββββββββββββ
print("\n" + "=" * 80)
print("5. REGION SCATTER PLOTS")
print("=" * 80)
fig, axes = plt.subplots(1, 3, figsize=(21, 7))
for i, region in enumerate(["conserved", "neutral", "accelerated"]):
ax = axes[i]
for name, color in COLORS.items():
key = (name, region)
if key in agg_region:
d = agg_region[key]
ax.scatter(d["mean_mean"], d["mean_var"], alpha=0.4, s=20, label=name, color=color)
ax.set_xlabel("Mean phyloP")
ax.set_ylabel("Mean Variance")
ax.set_title(f"{region.capitalize()} Regions")
ax.legend(fontsize=7, loc="upper left")
plt.suptitle("Region-Based: Ablation vs References", fontsize=14)
plt.tight_layout()
plt.savefig(os.path.join(PLOT_DIR, "scatter_by_region.png"), dpi=150, bbox_inches="tight")
plt.close()
print(f" Saved scatter_by_region.png")
# ββ 6. Bar charts: Summary metrics by tokenizer and region ββββββββββββββββββ
print("\n" + "=" * 80)
print("6. BAR CHARTS: Summary metrics")
print("=" * 80)
metrics = ["mean_of_mean_phyloP", "pct_mean_phyloP_above_0", "mean_of_variance", "pct_variance_below_0.1"]
metric_labels = ["Mean of Mean phyloP", "% Mean phyloP > 0", "Mean of Variance", "% Variance < 0.1"]
fig, axes = plt.subplots(2, 2, figsize=(18, 12))
for idx, (metric, label) in enumerate(zip(metrics, metric_labels)):
ax = axes[idx // 2, idx % 2]
for region in ["conserved", "neutral", "accelerated"]:
subset = df_region[df_region["region"] == region].copy()
subset = subset.sort_values("tokenizer")
x = range(len(subset))
offset = {"conserved": -0.25, "neutral": 0, "accelerated": 0.25}[region]
color = {"conserved": "forestgreen", "neutral": "gray", "accelerated": "firebrick"}[region]
ax.bar([xi + offset for xi in x], subset[metric], width=0.25,
label=region, color=color, alpha=0.8)
ax.set_xticks(range(len(subset)))
ax.set_xticklabels(subset["tokenizer"], rotation=45, ha="right", fontsize=8)
ax.set_ylabel(label)
ax.set_title(label)
ax.legend(fontsize=8)
plt.suptitle("Ablation: Region-Based Metrics", fontsize=14)
plt.tight_layout()
plt.savefig(os.path.join(PLOT_DIR, "bar_region_metrics.png"), dpi=150, bbox_inches="tight")
plt.close()
print(f" Saved bar_region_metrics.png")
# ββ 7. Grouped bar: Ablation comparison (window metrics) ββββββββββββββββββββ
print("\n" + "=" * 80)
print("7. GROUPED BAR: Window metrics comparison")
print("=" * 80)
order = ["Baseline_bpe_5120", "Ablation_no_partition", "Ablation_no_priority",
"Ablation_no_length", "Merged_uni_len2_5120"]
df_w_ordered = df_window.set_index("tokenizer").loc[order].reset_index()
fig, axes = plt.subplots(1, 3, figsize=(20, 6))
# Mean phyloP
ax = axes[0]
bars = ax.barh(df_w_ordered["tokenizer"], df_w_ordered["mean_of_mean_phyloP"],
color=[COLORS[n] for n in df_w_ordered["tokenizer"]])
ax.set_xlabel("Mean of Mean phyloP")
ax.set_title("Conservation Signal (Higher is Better)")
ax.invert_yaxis()
for bar, val in zip(bars, df_w_ordered["mean_of_mean_phyloP"]):
ax.text(bar.get_width() + 0.001, bar.get_y() + bar.get_height()/2,
f"{val:.4f}", va="center", fontsize=9)
# % above 0
ax = axes[1]
bars = ax.barh(df_w_ordered["tokenizer"], df_w_ordered["pct_mean_phyloP_above_0"],
color=[COLORS[n] for n in df_w_ordered["tokenizer"]])
ax.set_xlabel("% Mean phyloP > 0")
ax.set_title("Positive Conservation Rate (Higher is Better)")
ax.invert_yaxis()
for bar, val in zip(bars, df_w_ordered["pct_mean_phyloP_above_0"]):
ax.text(bar.get_width() + 0.1, bar.get_y() + bar.get_height()/2,
f"{val:.1f}%", va="center", fontsize=9)
# Mean variance
ax = axes[2]
bars = ax.barh(df_w_ordered["tokenizer"], df_w_ordered["mean_of_variance"],
color=[COLORS[n] for n in df_w_ordered["tokenizer"]])
ax.set_xlabel("Mean of Variance")
ax.set_title("Internal Variance (Lower is Better)")
ax.invert_yaxis()
for bar, val in zip(bars, df_w_ordered["mean_of_variance"]):
ax.text(bar.get_width() + 0.001, bar.get_y() + bar.get_height()/2,
f"{val:.4f}", va="center", fontsize=9)
plt.suptitle("Ablation: Window-Based Summary Metrics", fontsize=14)
plt.tight_layout()
plt.savefig(os.path.join(PLOT_DIR, "bar_window_comparison.png"), dpi=150, bbox_inches="tight")
plt.close()
print(f" Saved bar_window_comparison.png")
# ββ 8. Print final comparison table βββββββββββββββββββββββββββββββββββββββββ
print("\n" + "=" * 80)
print("8. FINAL COMPARISON TABLE")
print("=" * 80)
print("\nWindow-based:")
print(df_w_ordered[["tokenizer", "mean_of_mean_phyloP", "pct_mean_phyloP_above_0",
"mean_of_variance", "pct_variance_below_0.1", "num_tokens"]].to_string(index=False))
print("\nRegion-based (conserved only):")
cons = df_region[df_region["region"] == "conserved"].copy()
cons = cons[cons["tokenizer"].isin(order)]
print(cons[["tokenizer", "mean_of_mean_phyloP", "pct_mean_phyloP_above_0",
"mean_of_variance", "num_tokens"]].to_string(index=False))
print(f"\nAll plots saved to: {PLOT_DIR}/")
print("All summary CSVs saved to:", ABLATION_DIR)
|