File size: 8,647 Bytes
1393c62 | 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 | #!/usr/bin/env python3
"""
Comprehensive Cross-Dataset Evaluation for Thyroid Ultrasound Model
Computes: Accuracy, Sensitivity, Specificity, PPV, NPV, AUC-ROC, F1
Evaluates on:
1. BTX24 test split (same-dataset validation)
2. joooy94/thyroid_data (cross-dataset validation)
Results pushed to Hugging Face Hub.
"""
import os, sys, json, warnings, traceback
warnings.filterwarnings("ignore")
import numpy as np
from datasets import load_dataset
from transformers import AutoImageProcessor, AutoModelForImageClassification
from sklearn.metrics import (
accuracy_score, precision_score, recall_score, f1_score,
roc_auc_score, confusion_matrix, precision_recall_fscore_support
)
import torch
import torch.nn.functional as F
from huggingface_hub import HfApi
HF_USERNAME = "Johnyquest7"
MODEL_NAME = f"{HF_USERNAME}/ML-Inter_thyroid"
REPO_ID = f"{HF_USERNAME}/thyroid-training-scripts"
SEED = 42
BATCH_SIZE = 8 # Smaller for CPU compatibility
np.random.seed(SEED)
torch.manual_seed(SEED)
def evaluate_dataset(dataset_name, split_name, label_column, dataset_is_split=True):
"""Evaluate model on a dataset. Returns metrics dict."""
print(f"\n{'='*60}")
print(f"Evaluating on: {dataset_name} | split: {split_name}")
print(f"{'='*60}")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Device: {device}")
# Load model once
print(f"Loading model: {MODEL_NAME}")
processor = AutoImageProcessor.from_pretrained(MODEL_NAME)
model = AutoModelForImageClassification.from_pretrained(MODEL_NAME).to(device).eval()
id2label = model.config.id2label
print(f"Model classes: {id2label}")
# Load dataset
print(f"Loading dataset: {dataset_name}")
try:
if dataset_is_split:
ds = load_dataset(dataset_name, split=split_name)
else:
ds = load_dataset(dataset_name)
if split_name in ds:
ds = ds[split_name]
else:
ds = ds[list(ds.keys())[0]]
except Exception as e:
print(f"ERROR loading dataset: {e}")
return {"error": str(e)}
print(f"Total samples: {len(ds)}")
# Check if dataset has required columns
if "image" not in ds.column_names:
print(f"ERROR: Dataset missing 'image' column. Available: {ds.column_names}")
return {"error": "Missing image column"}
if label_column not in ds.column_names:
print(f"ERROR: Dataset missing '{label_column}' column. Available: {ds.column_names}")
return {"error": f"Missing {label_column} column"}
# Count labels
labels = [ds[i][label_column] for i in range(min(100, len(ds)))]
unique_labels = sorted(set(labels))
print(f"Label values (first 100): {unique_labels}")
# Map dataset labels to model labels if needed
# Assume 0 = benign, 1 = malignant (standard convention)
# If labels are different, we may need mapping
all_logits, all_labels = [], []
for i in range(0, len(ds), BATCH_SIZE):
batch_items = [ds[j] for j in range(i, min(i+BATCH_SIZE, len(ds)))]
try:
images = []
valid_labels = []
for item in batch_items:
img = item["image"]
if hasattr(img, 'mode'):
img = img.convert("RGB") if img.mode != "RGB" else img
elif hasattr(img, 'convert'):
img = img.convert("RGB")
images.append(img)
valid_labels.append(item[label_column])
inputs = processor(images, return_tensors="pt")
with torch.no_grad():
outputs = model(pixel_values=inputs["pixel_values"].to(device))
all_logits.extend(outputs.logits.cpu().numpy())
all_labels.extend(valid_labels)
except Exception as e:
print(f" Error in batch {i//BATCH_SIZE}: {e}")
continue
if (i // BATCH_SIZE) % 10 == 0:
print(f" Processed {i}/{len(ds)} samples")
print(f"\nTotal evaluated: {len(all_labels)}")
if len(all_labels) == 0:
return {"error": "No samples evaluated"}
y_true = np.array(all_labels)
y_logits = np.array(all_logits)
y_pred = np.argmax(y_logits, axis=1)
probs = F.softmax(torch.from_numpy(y_logits), dim=1).numpy()
# Compute all metrics
acc = accuracy_score(y_true, y_pred)
prec, rec, f1, _ = precision_recall_fscore_support(y_true, y_pred, average="weighted", zero_division=0)
# Binary metrics
cm = confusion_matrix(y_true, y_pred)
print(f"\nConfusion Matrix:\n{cm}")
# Handle different label conventions
# If dataset uses 0=benign, 1=malignant (same as model)
tn, fp, fn, tp = cm.ravel() if cm.size == 4 else (0, 0, 0, 0)
sensitivity = tp / (tp + fn) if (tp + fn) > 0 else 0.0
specificity = tn / (tn + fp) if (tn + fp) > 0 else 0.0
ppv = tp / (tp + fp) if (tp + fp) > 0 else 0.0
npv = tn / (tn + fn) if (tn + fn) > 0 else 0.0
# AUC-ROC
try:
if probs.shape[1] >= 2:
auc = roc_auc_score(y_true, probs[:, 1])
else:
auc = roc_auc_score(y_true, probs[:, 0])
except Exception as e:
print(f"AUC calculation failed: {e}")
auc = 0.0
# Per-class metrics
prec_macro = precision_score(y_true, y_pred, average="macro", zero_division=0)
rec_macro = recall_score(y_true, y_pred, average="macro", zero_division=0)
f1_macro = f1_score(y_true, y_pred, average="macro", zero_division=0)
metrics = {
"dataset": dataset_name,
"split": split_name,
"n_samples": int(len(y_true)),
"accuracy": float(acc),
"weighted_precision": float(prec),
"weighted_recall": float(rec),
"weighted_f1": float(f1),
"macro_precision": float(prec_macro),
"macro_recall": float(rec_macro),
"macro_f1": float(f1_macro),
"sensitivity": float(sensitivity),
"specificity": float(specificity),
"ppv": float(ppv),
"npv": float(npv),
"auc_roc": float(auc),
"confusion_matrix": cm.tolist(),
}
print(f"\n{'='*60}")
print("RESULTS")
print(f"{'='*60}")
for k, v in metrics.items():
if k != "confusion_matrix":
print(f" {k}: {v}")
return metrics
def main():
print("=" * 60)
print("Cross-Dataset Thyroid Model Evaluation")
print("=" * 60)
all_results = {}
# 1. Evaluate on BTX24 test split (our own held-out data)
try:
ds_full = load_dataset("BTX24/thyroid-cancer-classification-ultrasound-dataset", split="train")
ds_full = ds_full.shuffle(seed=SEED)
train_test = ds_full.train_test_split(test_size=0.2, stratify_by_column="label", seed=SEED)
test_ds = train_test["test"]
# Save test_ds as temporary and evaluate
print(f"\nBTX24 Test Split: {len(test_ds)} samples")
metrics_btx24 = evaluate_dataset(
"BTX24/thyroid-cancer-classification-ultrasound-dataset",
"train",
"label",
dataset_is_split=True
)
all_results["BTX24_test_split"] = metrics_btx24
except Exception as e:
print(f"BTX24 evaluation failed: {e}")
traceback.print_exc()
all_results["BTX24_test_split"] = {"error": str(e)}
# 2. Evaluate on joooy94/thyroid_data (cross-dataset)
try:
metrics_cross = evaluate_dataset(
"joooy94/thyroid_data",
"train",
"label",
dataset_is_split=True
)
all_results["joooy94_thyroid_data"] = metrics_cross
except Exception as e:
print(f"joooy94 evaluation failed: {e}")
traceback.print_exc()
all_results["joooy94_thyroid_data"] = {"error": str(e)}
# Save results
print(f"\n{'='*60}")
print("SAVING RESULTS")
print(f"{'='*60}")
results_json = json.dumps(all_results, indent=2)
print(results_json)
# Write to local file
output_path = "/tmp/cross_dataset_metrics.json"
with open(output_path, "w") as f:
f.write(results_json)
print(f"\nSaved to {output_path}")
# Upload to Hub
try:
api = HfApi()
api.upload_file(
path_or_fileobj=output_path,
path_in_file="cross_dataset_metrics.json",
repo_id=REPO_ID,
repo_type="model"
)
print(f"Uploaded to https://huggingface.co/{REPO_ID}/blob/main/cross_dataset_metrics.json")
except Exception as e:
print(f"Upload failed: {e}")
traceback.print_exc()
print("\nDone!")
if __name__ == "__main__":
main()
|