本实验主要测试每个因素叠加所使用的最佳权重维度，只有6个实验，分别用raw,32,64,128进行测试，全部数据的之前已经测过了，在128的时候最好¶
In [1]:
import os
import copy
import json
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
from torch.utils.data import DataLoader, Dataset
from torchvision import transforms
from torchvision.models import resnet50
from PIL import Image
from sklearn.metrics import (
accuracy_score,
balanced_accuracy_score,
f1_score
)
from sklearn.utils.class_weight import compute_class_weight
In [2]:
device = torch.device(
"mps" if torch.backends.mps.is_available()
else "cuda" if torch.cuda.is_available()
else "cpu"
)
print("Device:", device)
PROJECT_DIR = "/Users/applesues01/Documents/Medical_Agent"
DATA_DIR = os.path.join(PROJECT_DIR, "data", "HAM10000")
IMAGE_DIR1 = os.path.join(DATA_DIR, "HAM10000_images_part_1")
IMAGE_DIR2 = os.path.join(DATA_DIR, "HAM10000_images_part_2")
SPLIT_DIR = os.path.join(DATA_DIR, "splits")
CHECKPOINT_DIR = os.path.join(PROJECT_DIR, "checkpoints")
SUPPORT_DIR = os.path.join(PROJECT_DIR, "supports")
Device: mps
In [3]:
train_df = pd.read_csv(os.path.join(SPLIT_DIR, "train.csv"))
val_df = pd.read_csv(os.path.join(SPLIT_DIR, "val.csv"))
test_df = pd.read_csv(os.path.join(SPLIT_DIR, "test.csv"))
print(len(train_df), len(val_df), len(test_df))
7002 1532 1481
In [4]:
LABEL_MAP = {
"akiec": 0,
"bcc": 1,
"bkl": 2,
"df": 3,
"mel": 4,
"nv": 5,
"vasc": 6,
}
CLASS_NAMES = ["akiec", "bcc", "bkl", "df", "mel", "nv", "vasc"]
LOCATIONS = [
"scalp", "ear", "face", "back", "trunk", "chest",
"upper extremity", "abdomen", "unknown", "lower extremity",
"genital", "neck", "hand", "foot", "acral"
]
SEX_MAP = {
"male": [1.0, 0.0, 0.0],
"female": [0.0, 1.0, 0.0],
"unknown": [0.0, 0.0, 1.0],
}
train_age_mean = train_df["age"].mean()
def process_selected_metadata(row, selected_features, train_age_mean):
features = []
if "age" in selected_features:
age = row["age"]
if pd.isna(age):
age = train_age_mean
features.append(float(age) / 100.0)
if "sex" in selected_features:
sex_key = row["sex"] if row["sex"] in SEX_MAP else "unknown"
features.extend(SEX_MAP[sex_key])
if "location" in selected_features:
loc_key = row["localization"] if row["localization"] in LOCATIONS else "unknown"
loc_vector = [0.0] * len(LOCATIONS)
loc_vector[LOCATIONS.index(loc_key)] = 1.0
features.extend(loc_vector)
return np.array(features, dtype=np.float32)
In [5]:
metadata_experiments = [
{"name": "image_age", "features": ["age"]},
{"name": "image_sex", "features": ["sex"]},
{"name": "image_location", "features": ["location"]},
{"name": "image_age_sex", "features": ["age", "sex"]},
{"name": "image_age_location", "features": ["age", "location"]},
{"name": "image_sex_location", "features": ["sex", "location"]},
]
embed_dims_to_try = [32, 64, 128]
metadata_experiments
Out[5]:
[{'name': 'image_age', 'features': ['age']},
{'name': 'image_sex', 'features': ['sex']},
{'name': 'image_location', 'features': ['location']},
{'name': 'image_age_sex', 'features': ['age', 'sex']},
{'name': 'image_age_location', 'features': ['age', 'location']},
{'name': 'image_sex_location', 'features': ['sex', 'location']}]
In [6]:
IMAGE_SIZE = 224
BATCH_SIZE = 16
eval_transform = transforms.Compose([
transforms.Resize((IMAGE_SIZE, IMAGE_SIZE)),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]
)
])
def resolve_image_path(image_id):
filename = f"{image_id}.jpg"
path1 = os.path.join(IMAGE_DIR1, filename)
path2 = os.path.join(IMAGE_DIR2, filename)
return path1 if os.path.exists(path1) else path2
In [7]:
class HAMImageDataset(Dataset):
def __init__(self, dataframe, transform=None):
self.df = dataframe.reset_index(drop=True)
self.transform = transform
def __len__(self):
return len(self.df)
def __getitem__(self, idx):
row = self.df.iloc[idx]
image = Image.open(resolve_image_path(row["image_id"])).convert("RGB")
if self.transform:
image = self.transform(image)
label = LABEL_MAP[row["dx"]]
return image, label
In [8]:
image_only_model = resnet50(weights=None)
image_only_model.fc = nn.Linear(image_only_model.fc.in_features, 7)
checkpoint_path = os.path.join(
CHECKPOINT_DIR,
"resnet50_image_only_finetuned_best.pth"
)
image_only_model.load_state_dict(
torch.load(checkpoint_path, map_location=device)
)
image_only_model = image_only_model.to(device)
image_only_model.eval()
print("Image Only checkpoint loaded")
Image Only checkpoint loaded
In [9]:
class ResNet50FeatureExtractor(nn.Module):
def __init__(self, backbone):
super().__init__()
self.features = nn.Sequential(*list(backbone.children())[:-1])
def forward(self, x):
x = self.features(x)
x = torch.flatten(x, 1)
return x
In [10]:
feature_extractor = ResNet50FeatureExtractor(image_only_model).to(device)
feature_extractor.eval()
for param in feature_extractor.parameters():
param.requires_grad = False
print("Feature extractor ready")
Feature extractor ready
In [17]:
def extract_image_features(data_loader, feature_extractor, device):
feature_extractor.eval()
all_features = []
all_labels = []
with torch.no_grad():
for images, labels in data_loader:
images = images.to(device)
features = feature_extractor(images).cpu()
all_features.append(features)
all_labels.append(labels)
return torch.cat(all_features, dim=0), torch.cat(all_labels, dim=0)
In [13]:
train_image_loader = DataLoader(
HAMImageDataset(train_df, eval_transform),
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=0
)
val_image_loader = DataLoader(
HAMImageDataset(val_df, eval_transform),
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=0
)
test_image_loader = DataLoader(
HAMImageDataset(test_df, eval_transform),
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=0
)
train_image_features, train_labels = extract_image_features(
train_image_loader, feature_extractor, device
)
val_image_features, val_labels = extract_image_features(
val_image_loader, feature_extractor, device
)
test_image_features, test_labels = extract_image_features(
test_image_loader, feature_extractor, device
)
print(train_image_features.shape, train_labels.shape)
print(val_image_features.shape, val_labels.shape)
print(test_image_features.shape, test_labels.shape)
torch.Size([7002, 2048]) torch.Size([7002]) torch.Size([1532, 2048]) torch.Size([1532]) torch.Size([1481, 2048]) torch.Size([1481])
In [14]:
def build_metadata_matrix(dataframe, selected_features, train_age_mean):
rows = [
process_selected_metadata(
dataframe.iloc[idx],
selected_features,
train_age_mean
)
for idx in range(len(dataframe))
]
return np.stack(rows, axis=0)
In [15]:
class CachedFusionDataset(Dataset):
def __init__(self, image_features, metadata_features, labels):
self.image_features = image_features.float()
self.metadata_features = torch.tensor(metadata_features, dtype=torch.float32)
self.labels = labels.long()
def __len__(self):
return len(self.labels)
def __getitem__(self, idx):
return (
self.image_features[idx],
self.metadata_features[idx],
self.labels[idx]
)
In [16]:
def build_cached_loaders_for_experiment(selected_features):
train_metadata = build_metadata_matrix(
train_df,
selected_features,
train_age_mean
)
val_metadata = build_metadata_matrix(
val_df,
selected_features,
train_age_mean
)
test_metadata = build_metadata_matrix(
test_df,
selected_features,
train_age_mean
)
train_dataset = CachedFusionDataset(
train_image_features,
train_metadata,
train_labels
)
val_dataset = CachedFusionDataset(
val_image_features,
val_metadata,
val_labels
)
test_dataset = CachedFusionDataset(
test_image_features,
test_metadata,
test_labels
)
train_loader = DataLoader(
train_dataset,
batch_size=BATCH_SIZE,
shuffle=True,
num_workers=0
)
val_loader = DataLoader(
val_dataset,
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=0
)
test_loader = DataLoader(
test_dataset,
batch_size=BATCH_SIZE,
shuffle=False,
num_workers=0
)
return train_loader, val_loader, test_loader, train_metadata.shape[1]
In [18]:
class MetadataEncoderFusionClassifier(nn.Module):
def __init__(self, metadata_input_dim, metadata_embed_dim=64, num_classes=7):
super().__init__()
self.metadata_encoder = nn.Sequential(
nn.Linear(metadata_input_dim, metadata_embed_dim),
nn.BatchNorm1d(metadata_embed_dim),
nn.ReLU(),
nn.Dropout(0.2)
)
self.classifier = nn.Sequential(
nn.Linear(2048 + metadata_embed_dim, 512),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(512, 128),
nn.ReLU(),
nn.Linear(128, num_classes)
)
def forward(self, image_features, metadata):
metadata_features = self.metadata_encoder(metadata)
fused = torch.cat([image_features, metadata_features], dim=1)
return self.classifier(fused)
In [19]:
def evaluate_cached_model(model, data_loader, criterion, device):
model.eval()
preds = []
truths = []
total_loss = 0.0
with torch.no_grad():
for image_features, metadata, labels in data_loader:
image_features = image_features.to(device)
metadata = metadata.to(device)
labels = labels.to(device)
outputs = model(image_features, metadata)
loss = criterion(outputs, labels)
total_loss += loss.item() * image_features.size(0)
preds.extend(outputs.argmax(1).cpu().numpy())
truths.extend(labels.cpu().numpy())
return {
"loss": total_loss / len(data_loader.dataset),
"accuracy": accuracy_score(truths, preds),
"balanced_accuracy": balanced_accuracy_score(truths, preds),
"macro_f1": f1_score(truths, preds, average="macro", zero_division=0),
}
def train_cached_one_experiment(model, data_loader, criterion, optimizer, device):
model.train()
preds = []
truths = []
total_loss = 0.0
for image_features, metadata, labels in data_loader:
image_features = image_features.to(device)
metadata = metadata.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = model(image_features, metadata)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
total_loss += loss.item() * image_features.size(0)
preds.extend(outputs.argmax(1).detach().cpu().numpy())
truths.extend(labels.cpu().numpy())
return {
"loss": total_loss / len(data_loader.dataset),
"accuracy": accuracy_score(truths, preds),
"macro_f1": f1_score(truths, preds, average="macro", zero_division=0),
}
In [20]:
class_weights = compute_class_weight(
class_weight="balanced",
classes=np.array(CLASS_NAMES),
y=train_df["dx"]
)
class_weights = torch.tensor(
class_weights,
dtype=torch.float32
).to(device)
print(class_weights)
tensor([ 4.3491, 2.7330, 1.2924, 13.1617, 1.2857, 0.2138, 10.1039],
device='mps:0')
In [21]:
def run_metadata_encoder_experiment(
experiment_name,
selected_features,
metadata_embed_dim=64,
num_epochs=20
):
train_loader, val_loader, test_loader, metadata_dim = build_cached_loaders_for_experiment(
selected_features
)
model = MetadataEncoderFusionClassifier(
metadata_input_dim=metadata_dim,
metadata_embed_dim=metadata_embed_dim,
num_classes=7
).to(device)
criterion = nn.CrossEntropyLoss(weight=class_weights)
optimizer = torch.optim.Adam(
model.parameters(),
lr=1e-4
)
best_val_f1 = -1.0
best_state = None
history = []
for epoch in range(num_epochs):
train_metrics = train_cached_one_experiment(
model,
train_loader,
criterion,
optimizer,
device
)
val_metrics = evaluate_cached_model(
model,
val_loader,
criterion,
device
)
history.append({
"epoch": epoch + 1,
"train_loss": train_metrics["loss"],
"train_accuracy": train_metrics["accuracy"],
"train_macro_f1": train_metrics["macro_f1"],
"val_loss": val_metrics["loss"],
"val_accuracy": val_metrics["accuracy"],
"val_balanced_accuracy": val_metrics["balanced_accuracy"],
"val_macro_f1": val_metrics["macro_f1"],
})
print(
f"[{experiment_name}] "
f"epoch {epoch+1}/{num_epochs} | "
f"train_f1={train_metrics['macro_f1']:.4f} | "
f"val_f1={val_metrics['macro_f1']:.4f} | "
f"val_bal_acc={val_metrics['balanced_accuracy']:.4f}"
)
if val_metrics["macro_f1"] > best_val_f1:
best_val_f1 = val_metrics["macro_f1"]
best_state = copy.deepcopy(model.state_dict())
model.load_state_dict(best_state)
test_metrics = evaluate_cached_model(
model,
test_loader,
criterion,
device
)
return {
"name": experiment_name,
"features": selected_features,
"metadata_dim": metadata_dim,
"metadata_embed_dim": metadata_embed_dim,
"best_val_macro_f1": best_val_f1,
"test_accuracy": test_metrics["accuracy"],
"test_balanced_accuracy": test_metrics["balanced_accuracy"],
"test_macro_f1": test_metrics["macro_f1"],
"history": history
}
In [22]:
trial_result = run_metadata_encoder_experiment(
experiment_name="image_age_encoded32",
selected_features=["age"],
metadata_embed_dim=32,
num_epochs=5
)
trial_result
[image_age_encoded32] epoch 1/5 | train_f1=0.4757 | val_f1=0.4573 | val_bal_acc=0.5159 [image_age_encoded32] epoch 2/5 | train_f1=0.7009 | val_f1=0.5360 | val_bal_acc=0.6046 [image_age_encoded32] epoch 3/5 | train_f1=0.7846 | val_f1=0.5353 | val_bal_acc=0.5824 [image_age_encoded32] epoch 4/5 | train_f1=0.8265 | val_f1=0.5524 | val_bal_acc=0.5819 [image_age_encoded32] epoch 5/5 | train_f1=0.8549 | val_f1=0.5591 | val_bal_acc=0.5878
Out[22]:
{'name': 'image_age_encoded32',
'features': ['age'],
'metadata_dim': 1,
'metadata_embed_dim': 32,
'best_val_macro_f1': 0.5590734262658515,
'test_accuracy': 0.7643484132343011,
'test_balanced_accuracy': 0.6185011277441932,
'test_macro_f1': 0.5769331456378782,
'history': [{'epoch': 1,
'train_loss': 1.3757865863743934,
'train_accuracy': 0.7076549557269352,
'train_macro_f1': 0.4757486695259101,
'val_loss': 0.9205009663385137,
'val_accuracy': 0.6755874673629243,
'val_balanced_accuracy': 0.5158705515494769,
'val_macro_f1': 0.4572665622579147},
{'epoch': 2,
'train_loss': 0.6629017387771906,
'train_accuracy': 0.7853470437017995,
'train_macro_f1': 0.7008942334104724,
'val_loss': 0.811258202355462,
'val_accuracy': 0.6951697127937336,
'val_balanced_accuracy': 0.604645605174234,
'val_macro_f1': 0.5359549649618064},
{'epoch': 3,
'train_loss': 0.4142191265947783,
'train_accuracy': 0.8303341902313625,
'train_macro_f1': 0.7846178056249696,
'val_loss': 0.889071045594178,
'val_accuracy': 0.6690600522193212,
'val_balanced_accuracy': 0.5824252454612833,
'val_macro_f1': 0.5352937918190129},
{'epoch': 4,
'train_loss': 0.30858423042658295,
'train_accuracy': 0.8533276206798057,
'train_macro_f1': 0.8264624074773869,
'val_loss': 0.8315554425392698,
'val_accuracy': 0.6945169712793734,
'val_balanced_accuracy': 0.5819068126572706,
'val_macro_f1': 0.5523535341053204},
{'epoch': 5,
'train_loss': 0.24944227131153304,
'train_accuracy': 0.8727506426735219,
'train_macro_f1': 0.8549049486049007,
'val_loss': 0.7976379494715298,
'val_accuracy': 0.7271540469973891,
'val_balanced_accuracy': 0.5878216285337252,
'val_macro_f1': 0.5590734262658515}]}
In [23]:
# Cell A: 定义这次完整实验搜索空间
SEARCH_EXPERIMENTS = [
{"name": "image_age", "features": ["age"], "metadata_dim": 1},
{"name": "image_sex", "features": ["sex"], "metadata_dim": 3},
{"name": "image_location", "features": ["location"], "metadata_dim": 15},
{"name": "image_age_sex", "features": ["age", "sex"], "metadata_dim": 4},
{"name": "image_age_location", "features": ["age", "location"], "metadata_dim": 16},
{"name": "image_sex_location", "features": ["sex", "location"], "metadata_dim": 18},
]
EMBED_DIMS = [32, 64, 128]
SEARCH_EXPERIMENTS
Out[23]:
[{'name': 'image_age', 'features': ['age'], 'metadata_dim': 1},
{'name': 'image_sex', 'features': ['sex'], 'metadata_dim': 3},
{'name': 'image_location', 'features': ['location'], 'metadata_dim': 15},
{'name': 'image_age_sex', 'features': ['age', 'sex'], 'metadata_dim': 4},
{'name': 'image_age_location',
'features': ['age', 'location'],
'metadata_dim': 16},
{'name': 'image_sex_location',
'features': ['sex', 'location'],
'metadata_dim': 18}]
In [25]:
# Cell B: 一次性完整跑完
all_search_results = []
for exp in SEARCH_EXPERIMENTS:
for embed_dim in EMBED_DIMS:
exp_name = f"{exp['name']}_encoded{embed_dim}"
print(f"\n===== Running {exp_name} =====")
result = run_metadata_encoder_experiment(
experiment_name=exp_name,
selected_features=exp["features"],
metadata_embed_dim=embed_dim,
num_epochs=20 # 先5轮快速筛；如果你想更稳可以改10
)
all_search_results.append({
"method": exp_name,
"base_method": exp["name"],
"features": ",".join(exp["features"]),
"metadata_dim": exp["metadata_dim"],
"metadata_embed_dim": embed_dim,
"best_val_macro_f1": result["best_val_macro_f1"],
"test_accuracy": result["test_accuracy"],
"test_balanced_accuracy": result["test_balanced_accuracy"],
"test_macro_f1": result["test_macro_f1"],
"history": result["history"],
})
len(all_search_results)
===== Running image_age_encoded32 ===== [image_age_encoded32] epoch 1/20 | train_f1=0.4541 | val_f1=0.5271 | val_bal_acc=0.5556 [image_age_encoded32] epoch 2/20 | train_f1=0.7024 | val_f1=0.5259 | val_bal_acc=0.5920 [image_age_encoded32] epoch 3/20 | train_f1=0.7896 | val_f1=0.5261 | val_bal_acc=0.5875 [image_age_encoded32] epoch 4/20 | train_f1=0.8233 | val_f1=0.5572 | val_bal_acc=0.6171 [image_age_encoded32] epoch 5/20 | train_f1=0.8552 | val_f1=0.5665 | val_bal_acc=0.5910 [image_age_encoded32] epoch 6/20 | train_f1=0.8794 | val_f1=0.5601 | val_bal_acc=0.5931 [image_age_encoded32] epoch 7/20 | train_f1=0.9030 | val_f1=0.5734 | val_bal_acc=0.5858 [image_age_encoded32] epoch 8/20 | train_f1=0.9045 | val_f1=0.5734 | val_bal_acc=0.5886 [image_age_encoded32] epoch 9/20 | train_f1=0.9223 | val_f1=0.5687 | val_bal_acc=0.5662 [image_age_encoded32] epoch 10/20 | train_f1=0.9325 | val_f1=0.5707 | val_bal_acc=0.5719 [image_age_encoded32] epoch 11/20 | train_f1=0.9401 | val_f1=0.5746 | val_bal_acc=0.5792 [image_age_encoded32] epoch 12/20 | train_f1=0.9397 | val_f1=0.5832 | val_bal_acc=0.5769 [image_age_encoded32] epoch 13/20 | train_f1=0.9571 | val_f1=0.5826 | val_bal_acc=0.5827 [image_age_encoded32] epoch 14/20 | train_f1=0.9614 | val_f1=0.5899 | val_bal_acc=0.5929 [image_age_encoded32] epoch 15/20 | train_f1=0.9670 | val_f1=0.5635 | val_bal_acc=0.5396 [image_age_encoded32] epoch 16/20 | train_f1=0.9706 | val_f1=0.5925 | val_bal_acc=0.5799 [image_age_encoded32] epoch 17/20 | train_f1=0.9751 | val_f1=0.5779 | val_bal_acc=0.5765 [image_age_encoded32] epoch 18/20 | train_f1=0.9771 | val_f1=0.5636 | val_bal_acc=0.5782 [image_age_encoded32] epoch 19/20 | train_f1=0.9799 | val_f1=0.5770 | val_bal_acc=0.5709 [image_age_encoded32] epoch 20/20 | train_f1=0.9849 | val_f1=0.5786 | val_bal_acc=0.5702 ===== Running image_age_encoded64 ===== [image_age_encoded64] epoch 1/20 | train_f1=0.4880 | val_f1=0.4827 | val_bal_acc=0.6001 [image_age_encoded64] epoch 2/20 | train_f1=0.7025 | val_f1=0.5486 | val_bal_acc=0.5965 [image_age_encoded64] epoch 3/20 | train_f1=0.7846 | val_f1=0.5670 | val_bal_acc=0.5885 [image_age_encoded64] epoch 4/20 | train_f1=0.8173 | val_f1=0.5494 | val_bal_acc=0.5948 [image_age_encoded64] epoch 5/20 | train_f1=0.8534 | val_f1=0.5551 | val_bal_acc=0.5919 [image_age_encoded64] epoch 6/20 | train_f1=0.8818 | val_f1=0.5653 | val_bal_acc=0.5940 [image_age_encoded64] epoch 7/20 | train_f1=0.8920 | val_f1=0.5930 | val_bal_acc=0.6022 [image_age_encoded64] epoch 8/20 | train_f1=0.9047 | val_f1=0.5896 | val_bal_acc=0.5983 [image_age_encoded64] epoch 9/20 | train_f1=0.9243 | val_f1=0.5777 | val_bal_acc=0.5893 [image_age_encoded64] epoch 10/20 | train_f1=0.9302 | val_f1=0.5889 | val_bal_acc=0.5746 [image_age_encoded64] epoch 11/20 | train_f1=0.9426 | val_f1=0.5819 | val_bal_acc=0.5919 [image_age_encoded64] epoch 12/20 | train_f1=0.9467 | val_f1=0.5828 | val_bal_acc=0.5739 [image_age_encoded64] epoch 13/20 | train_f1=0.9515 | val_f1=0.5797 | val_bal_acc=0.5744 [image_age_encoded64] epoch 14/20 | train_f1=0.9596 | val_f1=0.5736 | val_bal_acc=0.5709 [image_age_encoded64] epoch 15/20 | train_f1=0.9683 | val_f1=0.5768 | val_bal_acc=0.5537 [image_age_encoded64] epoch 16/20 | train_f1=0.9723 | val_f1=0.5789 | val_bal_acc=0.5704 [image_age_encoded64] epoch 17/20 | train_f1=0.9749 | val_f1=0.5804 | val_bal_acc=0.5758 [image_age_encoded64] epoch 18/20 | train_f1=0.9746 | val_f1=0.5930 | val_bal_acc=0.5802 [image_age_encoded64] epoch 19/20 | train_f1=0.9802 | val_f1=0.5858 | val_bal_acc=0.5545 [image_age_encoded64] epoch 20/20 | train_f1=0.9843 | val_f1=0.5802 | val_bal_acc=0.5650 ===== Running image_age_encoded128 ===== [image_age_encoded128] epoch 1/20 | train_f1=0.4682 | val_f1=0.4859 | val_bal_acc=0.5945 [image_age_encoded128] epoch 2/20 | train_f1=0.7088 | val_f1=0.5272 | val_bal_acc=0.6037 [image_age_encoded128] epoch 3/20 | train_f1=0.7794 | val_f1=0.5501 | val_bal_acc=0.6021 [image_age_encoded128] epoch 4/20 | train_f1=0.8314 | val_f1=0.5784 | val_bal_acc=0.6013 [image_age_encoded128] epoch 5/20 | train_f1=0.8519 | val_f1=0.5756 | val_bal_acc=0.6042 [image_age_encoded128] epoch 6/20 | train_f1=0.8798 | val_f1=0.5926 | val_bal_acc=0.6113 [image_age_encoded128] epoch 7/20 | train_f1=0.8953 | val_f1=0.5824 | val_bal_acc=0.5810 [image_age_encoded128] epoch 8/20 | train_f1=0.9025 | val_f1=0.5630 | val_bal_acc=0.5984 [image_age_encoded128] epoch 9/20 | train_f1=0.9178 | val_f1=0.5610 | val_bal_acc=0.5808 [image_age_encoded128] epoch 10/20 | train_f1=0.9317 | val_f1=0.5796 | val_bal_acc=0.5738 [image_age_encoded128] epoch 11/20 | train_f1=0.9469 | val_f1=0.5919 | val_bal_acc=0.6007 [image_age_encoded128] epoch 12/20 | train_f1=0.9511 | val_f1=0.5767 | val_bal_acc=0.5925 [image_age_encoded128] epoch 13/20 | train_f1=0.9531 | val_f1=0.5833 | val_bal_acc=0.5697 [image_age_encoded128] epoch 14/20 | train_f1=0.9588 | val_f1=0.5776 | val_bal_acc=0.5841 [image_age_encoded128] epoch 15/20 | train_f1=0.9653 | val_f1=0.5810 | val_bal_acc=0.5664 [image_age_encoded128] epoch 16/20 | train_f1=0.9714 | val_f1=0.5880 | val_bal_acc=0.5827 [image_age_encoded128] epoch 17/20 | train_f1=0.9707 | val_f1=0.5876 | val_bal_acc=0.5621 [image_age_encoded128] epoch 18/20 | train_f1=0.9767 | val_f1=0.5661 | val_bal_acc=0.5614 [image_age_encoded128] epoch 19/20 | train_f1=0.9808 | val_f1=0.5833 | val_bal_acc=0.5631 [image_age_encoded128] epoch 20/20 | train_f1=0.9840 | val_f1=0.5944 | val_bal_acc=0.5836 ===== Running image_sex_encoded32 ===== [image_sex_encoded32] epoch 1/20 | train_f1=0.4535 | val_f1=0.4711 | val_bal_acc=0.5526 [image_sex_encoded32] epoch 2/20 | train_f1=0.6985 | val_f1=0.4981 | val_bal_acc=0.5877 [image_sex_encoded32] epoch 3/20 | train_f1=0.7783 | val_f1=0.5326 | val_bal_acc=0.6021 [image_sex_encoded32] epoch 4/20 | train_f1=0.8187 | val_f1=0.5524 | val_bal_acc=0.6044 [image_sex_encoded32] epoch 5/20 | train_f1=0.8498 | val_f1=0.5500 | val_bal_acc=0.5601 [image_sex_encoded32] epoch 6/20 | train_f1=0.8754 | val_f1=0.5804 | val_bal_acc=0.5999 [image_sex_encoded32] epoch 7/20 | train_f1=0.8969 | val_f1=0.5740 | val_bal_acc=0.5953 [image_sex_encoded32] epoch 8/20 | train_f1=0.9019 | val_f1=0.5963 | val_bal_acc=0.5900 [image_sex_encoded32] epoch 9/20 | train_f1=0.9200 | val_f1=0.5802 | val_bal_acc=0.5861 [image_sex_encoded32] epoch 10/20 | train_f1=0.9343 | val_f1=0.5700 | val_bal_acc=0.5778 [image_sex_encoded32] epoch 11/20 | train_f1=0.9420 | val_f1=0.5676 | val_bal_acc=0.5785 [image_sex_encoded32] epoch 12/20 | train_f1=0.9463 | val_f1=0.5851 | val_bal_acc=0.5635 [image_sex_encoded32] epoch 13/20 | train_f1=0.9549 | val_f1=0.5831 | val_bal_acc=0.5786 [image_sex_encoded32] epoch 14/20 | train_f1=0.9638 | val_f1=0.5859 | val_bal_acc=0.5730 [image_sex_encoded32] epoch 15/20 | train_f1=0.9667 | val_f1=0.5742 | val_bal_acc=0.5551 [image_sex_encoded32] epoch 16/20 | train_f1=0.9728 | val_f1=0.5777 | val_bal_acc=0.5670 [image_sex_encoded32] epoch 17/20 | train_f1=0.9799 | val_f1=0.5800 | val_bal_acc=0.5602 [image_sex_encoded32] epoch 18/20 | train_f1=0.9801 | val_f1=0.5814 | val_bal_acc=0.5573 [image_sex_encoded32] epoch 19/20 | train_f1=0.9804 | val_f1=0.5776 | val_bal_acc=0.5723 [image_sex_encoded32] epoch 20/20 | train_f1=0.9863 | val_f1=0.5761 | val_bal_acc=0.5844 ===== Running image_sex_encoded64 ===== [image_sex_encoded64] epoch 1/20 | train_f1=0.5009 | val_f1=0.4709 | val_bal_acc=0.5811 [image_sex_encoded64] epoch 2/20 | train_f1=0.7024 | val_f1=0.5091 | val_bal_acc=0.6132 [image_sex_encoded64] epoch 3/20 | train_f1=0.7710 | val_f1=0.5399 | val_bal_acc=0.6015 [image_sex_encoded64] epoch 4/20 | train_f1=0.8254 | val_f1=0.5515 | val_bal_acc=0.6053 [image_sex_encoded64] epoch 5/20 | train_f1=0.8545 | val_f1=0.5712 | val_bal_acc=0.5978 [image_sex_encoded64] epoch 6/20 | train_f1=0.8806 | val_f1=0.5726 | val_bal_acc=0.5973 [image_sex_encoded64] epoch 7/20 | train_f1=0.8970 | val_f1=0.5844 | val_bal_acc=0.6056 [image_sex_encoded64] epoch 8/20 | train_f1=0.9071 | val_f1=0.5847 | val_bal_acc=0.5861 [image_sex_encoded64] epoch 9/20 | train_f1=0.9180 | val_f1=0.6009 | val_bal_acc=0.5900 [image_sex_encoded64] epoch 10/20 | train_f1=0.9278 | val_f1=0.5859 | val_bal_acc=0.5916 [image_sex_encoded64] epoch 11/20 | train_f1=0.9429 | val_f1=0.5631 | val_bal_acc=0.5952 [image_sex_encoded64] epoch 12/20 | train_f1=0.9490 | val_f1=0.5943 | val_bal_acc=0.5885 [image_sex_encoded64] epoch 13/20 | train_f1=0.9585 | val_f1=0.5644 | val_bal_acc=0.5686 [image_sex_encoded64] epoch 14/20 | train_f1=0.9571 | val_f1=0.5921 | val_bal_acc=0.5863 [image_sex_encoded64] epoch 15/20 | train_f1=0.9671 | val_f1=0.5722 | val_bal_acc=0.5923 [image_sex_encoded64] epoch 16/20 | train_f1=0.9667 | val_f1=0.5869 | val_bal_acc=0.5664 [image_sex_encoded64] epoch 17/20 | train_f1=0.9750 | val_f1=0.5937 | val_bal_acc=0.5841 [image_sex_encoded64] epoch 18/20 | train_f1=0.9779 | val_f1=0.5857 | val_bal_acc=0.5595 [image_sex_encoded64] epoch 19/20 | train_f1=0.9767 | val_f1=0.5762 | val_bal_acc=0.5619 [image_sex_encoded64] epoch 20/20 | train_f1=0.9826 | val_f1=0.5865 | val_bal_acc=0.5608 ===== Running image_sex_encoded128 ===== [image_sex_encoded128] epoch 1/20 | train_f1=0.4794 | val_f1=0.4701 | val_bal_acc=0.5340 [image_sex_encoded128] epoch 2/20 | train_f1=0.6931 | val_f1=0.5186 | val_bal_acc=0.5833 [image_sex_encoded128] epoch 3/20 | train_f1=0.7738 | val_f1=0.5780 | val_bal_acc=0.6091 [image_sex_encoded128] epoch 4/20 | train_f1=0.8119 | val_f1=0.5794 | val_bal_acc=0.5973 [image_sex_encoded128] epoch 5/20 | train_f1=0.8381 | val_f1=0.5595 | val_bal_acc=0.6125 [image_sex_encoded128] epoch 6/20 | train_f1=0.8702 | val_f1=0.5840 | val_bal_acc=0.5837 [image_sex_encoded128] epoch 7/20 | train_f1=0.8939 | val_f1=0.5753 | val_bal_acc=0.5864 [image_sex_encoded128] epoch 8/20 | train_f1=0.8988 | val_f1=0.5803 | val_bal_acc=0.5885 [image_sex_encoded128] epoch 9/20 | train_f1=0.9153 | val_f1=0.5893 | val_bal_acc=0.5808 [image_sex_encoded128] epoch 10/20 | train_f1=0.9173 | val_f1=0.5699 | val_bal_acc=0.5659 [image_sex_encoded128] epoch 11/20 | train_f1=0.9357 | val_f1=0.5893 | val_bal_acc=0.5705 [image_sex_encoded128] epoch 12/20 | train_f1=0.9464 | val_f1=0.5834 | val_bal_acc=0.5542 [image_sex_encoded128] epoch 13/20 | train_f1=0.9517 | val_f1=0.5717 | val_bal_acc=0.5743 [image_sex_encoded128] epoch 14/20 | train_f1=0.9572 | val_f1=0.5844 | val_bal_acc=0.5837 [image_sex_encoded128] epoch 15/20 | train_f1=0.9623 | val_f1=0.5908 | val_bal_acc=0.5916 [image_sex_encoded128] epoch 16/20 | train_f1=0.9661 | val_f1=0.5919 | val_bal_acc=0.5820 [image_sex_encoded128] epoch 17/20 | train_f1=0.9697 | val_f1=0.5994 | val_bal_acc=0.5952 [image_sex_encoded128] epoch 18/20 | train_f1=0.9724 | val_f1=0.5588 | val_bal_acc=0.5797 [image_sex_encoded128] epoch 19/20 | train_f1=0.9624 | val_f1=0.5802 | val_bal_acc=0.5632 [image_sex_encoded128] epoch 20/20 | train_f1=0.9738 | val_f1=0.5999 | val_bal_acc=0.5940 ===== Running image_location_encoded32 ===== [image_location_encoded32] epoch 1/20 | train_f1=0.4579 | val_f1=0.4864 | val_bal_acc=0.6128 [image_location_encoded32] epoch 2/20 | train_f1=0.7034 | val_f1=0.5277 | val_bal_acc=0.5757 [image_location_encoded32] epoch 3/20 | train_f1=0.7886 | val_f1=0.5330 | val_bal_acc=0.6059 [image_location_encoded32] epoch 4/20 | train_f1=0.8299 | val_f1=0.5726 | val_bal_acc=0.6025 [image_location_encoded32] epoch 5/20 | train_f1=0.8606 | val_f1=0.5744 | val_bal_acc=0.5952 [image_location_encoded32] epoch 6/20 | train_f1=0.8896 | val_f1=0.5594 | val_bal_acc=0.5920 [image_location_encoded32] epoch 7/20 | train_f1=0.8974 | val_f1=0.5770 | val_bal_acc=0.5920 [image_location_encoded32] epoch 8/20 | train_f1=0.9136 | val_f1=0.5857 | val_bal_acc=0.5757 [image_location_encoded32] epoch 9/20 | train_f1=0.9200 | val_f1=0.5958 | val_bal_acc=0.5645 [image_location_encoded32] epoch 10/20 | train_f1=0.9315 | val_f1=0.5734 | val_bal_acc=0.5721 [image_location_encoded32] epoch 11/20 | train_f1=0.9447 | val_f1=0.5773 | val_bal_acc=0.5868 [image_location_encoded32] epoch 12/20 | train_f1=0.9520 | val_f1=0.5745 | val_bal_acc=0.5726 [image_location_encoded32] epoch 13/20 | train_f1=0.9586 | val_f1=0.5862 | val_bal_acc=0.5731 [image_location_encoded32] epoch 14/20 | train_f1=0.9671 | val_f1=0.6038 | val_bal_acc=0.5840 [image_location_encoded32] epoch 15/20 | train_f1=0.9700 | val_f1=0.5913 | val_bal_acc=0.5850 [image_location_encoded32] epoch 16/20 | train_f1=0.9749 | val_f1=0.5884 | val_bal_acc=0.5727 [image_location_encoded32] epoch 17/20 | train_f1=0.9787 | val_f1=0.5928 | val_bal_acc=0.5926 [image_location_encoded32] epoch 18/20 | train_f1=0.9826 | val_f1=0.6032 | val_bal_acc=0.5780 [image_location_encoded32] epoch 19/20 | train_f1=0.9858 | val_f1=0.5924 | val_bal_acc=0.5864 [image_location_encoded32] epoch 20/20 | train_f1=0.9827 | val_f1=0.5973 | val_bal_acc=0.5947 ===== Running image_location_encoded64 ===== [image_location_encoded64] epoch 1/20 | train_f1=0.4689 | val_f1=0.5023 | val_bal_acc=0.6010 [image_location_encoded64] epoch 2/20 | train_f1=0.7020 | val_f1=0.5531 | val_bal_acc=0.6163 [image_location_encoded64] epoch 3/20 | train_f1=0.7709 | val_f1=0.5425 | val_bal_acc=0.5960 [image_location_encoded64] epoch 4/20 | train_f1=0.8226 | val_f1=0.5790 | val_bal_acc=0.6052 [image_location_encoded64] epoch 5/20 | train_f1=0.8481 | val_f1=0.5676 | val_bal_acc=0.6002 [image_location_encoded64] epoch 6/20 | train_f1=0.8720 | val_f1=0.5493 | val_bal_acc=0.5789 [image_location_encoded64] epoch 7/20 | train_f1=0.8942 | val_f1=0.5855 | val_bal_acc=0.5978 [image_location_encoded64] epoch 8/20 | train_f1=0.9134 | val_f1=0.5829 | val_bal_acc=0.6007 [image_location_encoded64] epoch 9/20 | train_f1=0.9224 | val_f1=0.5784 | val_bal_acc=0.6040 [image_location_encoded64] epoch 10/20 | train_f1=0.9282 | val_f1=0.6006 | val_bal_acc=0.5873 [image_location_encoded64] epoch 11/20 | train_f1=0.9417 | val_f1=0.5885 | val_bal_acc=0.5741 [image_location_encoded64] epoch 12/20 | train_f1=0.9433 | val_f1=0.5863 | val_bal_acc=0.5949 [image_location_encoded64] epoch 13/20 | train_f1=0.9493 | val_f1=0.5943 | val_bal_acc=0.5793 [image_location_encoded64] epoch 14/20 | train_f1=0.9616 | val_f1=0.5922 | val_bal_acc=0.5690 [image_location_encoded64] epoch 15/20 | train_f1=0.9647 | val_f1=0.5855 | val_bal_acc=0.5760 [image_location_encoded64] epoch 16/20 | train_f1=0.9672 | val_f1=0.5903 | val_bal_acc=0.5563 [image_location_encoded64] epoch 17/20 | train_f1=0.9739 | val_f1=0.5879 | val_bal_acc=0.5783 [image_location_encoded64] epoch 18/20 | train_f1=0.9784 | val_f1=0.5957 | val_bal_acc=0.5739 [image_location_encoded64] epoch 19/20 | train_f1=0.9808 | val_f1=0.5925 | val_bal_acc=0.5883 [image_location_encoded64] epoch 20/20 | train_f1=0.9850 | val_f1=0.5933 | val_bal_acc=0.5890 ===== Running image_location_encoded128 ===== [image_location_encoded128] epoch 1/20 | train_f1=0.4894 | val_f1=0.5122 | val_bal_acc=0.5849 [image_location_encoded128] epoch 2/20 | train_f1=0.6916 | val_f1=0.5155 | val_bal_acc=0.5876 [image_location_encoded128] epoch 3/20 | train_f1=0.7741 | val_f1=0.5798 | val_bal_acc=0.6252 [image_location_encoded128] epoch 4/20 | train_f1=0.8169 | val_f1=0.5835 | val_bal_acc=0.6057 [image_location_encoded128] epoch 5/20 | train_f1=0.8525 | val_f1=0.5923 | val_bal_acc=0.5972 [image_location_encoded128] epoch 6/20 | train_f1=0.8746 | val_f1=0.5501 | val_bal_acc=0.5958 [image_location_encoded128] epoch 7/20 | train_f1=0.8901 | val_f1=0.5672 | val_bal_acc=0.5917 [image_location_encoded128] epoch 8/20 | train_f1=0.9090 | val_f1=0.5770 | val_bal_acc=0.5991 [image_location_encoded128] epoch 9/20 | train_f1=0.9166 | val_f1=0.5794 | val_bal_acc=0.5811 [image_location_encoded128] epoch 10/20 | train_f1=0.9279 | val_f1=0.5913 | val_bal_acc=0.5945 [image_location_encoded128] epoch 11/20 | train_f1=0.9375 | val_f1=0.5728 | val_bal_acc=0.5748 [image_location_encoded128] epoch 12/20 | train_f1=0.9438 | val_f1=0.5598 | val_bal_acc=0.5897 [image_location_encoded128] epoch 13/20 | train_f1=0.9534 | val_f1=0.5780 | val_bal_acc=0.5516 [image_location_encoded128] epoch 14/20 | train_f1=0.9612 | val_f1=0.5808 | val_bal_acc=0.5656 [image_location_encoded128] epoch 15/20 | train_f1=0.9623 | val_f1=0.5882 | val_bal_acc=0.5775 [image_location_encoded128] epoch 16/20 | train_f1=0.9704 | val_f1=0.5852 | val_bal_acc=0.5885 [image_location_encoded128] epoch 17/20 | train_f1=0.9728 | val_f1=0.5956 | val_bal_acc=0.5682 [image_location_encoded128] epoch 18/20 | train_f1=0.9761 | val_f1=0.5926 | val_bal_acc=0.5742 [image_location_encoded128] epoch 19/20 | train_f1=0.9788 | val_f1=0.5784 | val_bal_acc=0.5669 [image_location_encoded128] epoch 20/20 | train_f1=0.9847 | val_f1=0.5879 | val_bal_acc=0.5701 ===== Running image_age_sex_encoded32 ===== [image_age_sex_encoded32] epoch 1/20 | train_f1=0.5106 | val_f1=0.5170 | val_bal_acc=0.5650 [image_age_sex_encoded32] epoch 2/20 | train_f1=0.7100 | val_f1=0.5387 | val_bal_acc=0.6245 [image_age_sex_encoded32] epoch 3/20 | train_f1=0.7831 | val_f1=0.5449 | val_bal_acc=0.5966 [image_age_sex_encoded32] epoch 4/20 | train_f1=0.8323 | val_f1=0.5759 | val_bal_acc=0.5947 [image_age_sex_encoded32] epoch 5/20 | train_f1=0.8732 | val_f1=0.5494 | val_bal_acc=0.5988 [image_age_sex_encoded32] epoch 6/20 | train_f1=0.8767 | val_f1=0.5577 | val_bal_acc=0.5767 [image_age_sex_encoded32] epoch 7/20 | train_f1=0.9000 | val_f1=0.5880 | val_bal_acc=0.6048 [image_age_sex_encoded32] epoch 8/20 | train_f1=0.9113 | val_f1=0.5778 | val_bal_acc=0.5858 [image_age_sex_encoded32] epoch 9/20 | train_f1=0.9225 | val_f1=0.5724 | val_bal_acc=0.6044 [image_age_sex_encoded32] epoch 10/20 | train_f1=0.9326 | val_f1=0.5894 | val_bal_acc=0.5692 [image_age_sex_encoded32] epoch 11/20 | train_f1=0.9364 | val_f1=0.6031 | val_bal_acc=0.5931 [image_age_sex_encoded32] epoch 12/20 | train_f1=0.9478 | val_f1=0.5767 | val_bal_acc=0.5517 [image_age_sex_encoded32] epoch 13/20 | train_f1=0.9563 | val_f1=0.5955 | val_bal_acc=0.5801 [image_age_sex_encoded32] epoch 14/20 | train_f1=0.9615 | val_f1=0.5855 | val_bal_acc=0.5814 [image_age_sex_encoded32] epoch 15/20 | train_f1=0.9666 | val_f1=0.5931 | val_bal_acc=0.5863 [image_age_sex_encoded32] epoch 16/20 | train_f1=0.9709 | val_f1=0.5859 | val_bal_acc=0.5690 [image_age_sex_encoded32] epoch 17/20 | train_f1=0.9750 | val_f1=0.5712 | val_bal_acc=0.5591 [image_age_sex_encoded32] epoch 18/20 | train_f1=0.9767 | val_f1=0.5693 | val_bal_acc=0.5451 [image_age_sex_encoded32] epoch 19/20 | train_f1=0.9798 | val_f1=0.5989 | val_bal_acc=0.5760 [image_age_sex_encoded32] epoch 20/20 | train_f1=0.9830 | val_f1=0.5870 | val_bal_acc=0.5670 ===== Running image_age_sex_encoded64 ===== [image_age_sex_encoded64] epoch 1/20 | train_f1=0.4740 | val_f1=0.5141 | val_bal_acc=0.5560 [image_age_sex_encoded64] epoch 2/20 | train_f1=0.7063 | val_f1=0.5257 | val_bal_acc=0.6104 [image_age_sex_encoded64] epoch 3/20 | train_f1=0.7723 | val_f1=0.5421 | val_bal_acc=0.6029 [image_age_sex_encoded64] epoch 4/20 | train_f1=0.8282 | val_f1=0.5427 | val_bal_acc=0.6004 [image_age_sex_encoded64] epoch 5/20 | train_f1=0.8543 | val_f1=0.5758 | val_bal_acc=0.6038 [image_age_sex_encoded64] epoch 6/20 | train_f1=0.8735 | val_f1=0.5831 | val_bal_acc=0.6076 [image_age_sex_encoded64] epoch 7/20 | train_f1=0.8950 | val_f1=0.5957 | val_bal_acc=0.5852 [image_age_sex_encoded64] epoch 8/20 | train_f1=0.9070 | val_f1=0.6040 | val_bal_acc=0.5850 [image_age_sex_encoded64] epoch 9/20 | train_f1=0.9194 | val_f1=0.5712 | val_bal_acc=0.5728 [image_age_sex_encoded64] epoch 10/20 | train_f1=0.9245 | val_f1=0.5914 | val_bal_acc=0.5796 [image_age_sex_encoded64] epoch 11/20 | train_f1=0.9367 | val_f1=0.5870 | val_bal_acc=0.5915 [image_age_sex_encoded64] epoch 12/20 | train_f1=0.9527 | val_f1=0.5960 | val_bal_acc=0.5956 [image_age_sex_encoded64] epoch 13/20 | train_f1=0.9566 | val_f1=0.5962 | val_bal_acc=0.5964 [image_age_sex_encoded64] epoch 14/20 | train_f1=0.9642 | val_f1=0.5969 | val_bal_acc=0.5912 [image_age_sex_encoded64] epoch 15/20 | train_f1=0.9639 | val_f1=0.5901 | val_bal_acc=0.5904 [image_age_sex_encoded64] epoch 16/20 | train_f1=0.9678 | val_f1=0.5845 | val_bal_acc=0.5668 [image_age_sex_encoded64] epoch 17/20 | train_f1=0.9719 | val_f1=0.5647 | val_bal_acc=0.5655 [image_age_sex_encoded64] epoch 18/20 | train_f1=0.9736 | val_f1=0.5411 | val_bal_acc=0.6115 [image_age_sex_encoded64] epoch 19/20 | train_f1=0.9765 | val_f1=0.5793 | val_bal_acc=0.5673 [image_age_sex_encoded64] epoch 20/20 | train_f1=0.9818 | val_f1=0.5937 | val_bal_acc=0.5897 ===== Running image_age_sex_encoded128 ===== [image_age_sex_encoded128] epoch 1/20 | train_f1=0.4801 | val_f1=0.4978 | val_bal_acc=0.6072 [image_age_sex_encoded128] epoch 2/20 | train_f1=0.6938 | val_f1=0.5327 | val_bal_acc=0.6106 [image_age_sex_encoded128] epoch 3/20 | train_f1=0.7719 | val_f1=0.5367 | val_bal_acc=0.6211 [image_age_sex_encoded128] epoch 4/20 | train_f1=0.8226 | val_f1=0.5425 | val_bal_acc=0.6273 [image_age_sex_encoded128] epoch 5/20 | train_f1=0.8493 | val_f1=0.5425 | val_bal_acc=0.5955 [image_age_sex_encoded128] epoch 6/20 | train_f1=0.8802 | val_f1=0.5931 | val_bal_acc=0.5965 [image_age_sex_encoded128] epoch 7/20 | train_f1=0.8971 | val_f1=0.5638 | val_bal_acc=0.5864 [image_age_sex_encoded128] epoch 8/20 | train_f1=0.9075 | val_f1=0.5811 | val_bal_acc=0.5819 [image_age_sex_encoded128] epoch 9/20 | train_f1=0.9127 | val_f1=0.6021 | val_bal_acc=0.6025 [image_age_sex_encoded128] epoch 10/20 | train_f1=0.9221 | val_f1=0.5754 | val_bal_acc=0.5805 [image_age_sex_encoded128] epoch 11/20 | train_f1=0.9412 | val_f1=0.5932 | val_bal_acc=0.5932 [image_age_sex_encoded128] epoch 12/20 | train_f1=0.9450 | val_f1=0.5804 | val_bal_acc=0.5765 [image_age_sex_encoded128] epoch 13/20 | train_f1=0.9495 | val_f1=0.5794 | val_bal_acc=0.5848 [image_age_sex_encoded128] epoch 14/20 | train_f1=0.9576 | val_f1=0.5881 | val_bal_acc=0.5807 [image_age_sex_encoded128] epoch 15/20 | train_f1=0.9670 | val_f1=0.5895 | val_bal_acc=0.5721 [image_age_sex_encoded128] epoch 16/20 | train_f1=0.9673 | val_f1=0.5905 | val_bal_acc=0.5898 [image_age_sex_encoded128] epoch 17/20 | train_f1=0.9723 | val_f1=0.6027 | val_bal_acc=0.5906 [image_age_sex_encoded128] epoch 18/20 | train_f1=0.9715 | val_f1=0.5853 | val_bal_acc=0.5653 [image_age_sex_encoded128] epoch 19/20 | train_f1=0.9742 | val_f1=0.5792 | val_bal_acc=0.5741 [image_age_sex_encoded128] epoch 20/20 | train_f1=0.9809 | val_f1=0.5754 | val_bal_acc=0.5677 ===== Running image_age_location_encoded32 ===== [image_age_location_encoded32] epoch 1/20 | train_f1=0.4830 | val_f1=0.5283 | val_bal_acc=0.6027 [image_age_location_encoded32] epoch 2/20 | train_f1=0.7254 | val_f1=0.5244 | val_bal_acc=0.5915 [image_age_location_encoded32] epoch 3/20 | train_f1=0.7870 | val_f1=0.5626 | val_bal_acc=0.5977 [image_age_location_encoded32] epoch 4/20 | train_f1=0.8391 | val_f1=0.5450 | val_bal_acc=0.5803 [image_age_location_encoded32] epoch 5/20 | train_f1=0.8640 | val_f1=0.5738 | val_bal_acc=0.5906 [image_age_location_encoded32] epoch 6/20 | train_f1=0.8884 | val_f1=0.5906 | val_bal_acc=0.6047 [image_age_location_encoded32] epoch 7/20 | train_f1=0.8955 | val_f1=0.5892 | val_bal_acc=0.6056 [image_age_location_encoded32] epoch 8/20 | train_f1=0.9128 | val_f1=0.6022 | val_bal_acc=0.6066 [image_age_location_encoded32] epoch 9/20 | train_f1=0.9291 | val_f1=0.5892 | val_bal_acc=0.5852 [image_age_location_encoded32] epoch 10/20 | train_f1=0.9377 | val_f1=0.5875 | val_bal_acc=0.5632 [image_age_location_encoded32] epoch 11/20 | train_f1=0.9441 | val_f1=0.5891 | val_bal_acc=0.5823 [image_age_location_encoded32] epoch 12/20 | train_f1=0.9565 | val_f1=0.6031 | val_bal_acc=0.5872 [image_age_location_encoded32] epoch 13/20 | train_f1=0.9635 | val_f1=0.5898 | val_bal_acc=0.5917 [image_age_location_encoded32] epoch 14/20 | train_f1=0.9625 | val_f1=0.6045 | val_bal_acc=0.5851 [image_age_location_encoded32] epoch 15/20 | train_f1=0.9676 | val_f1=0.5650 | val_bal_acc=0.5921 [image_age_location_encoded32] epoch 16/20 | train_f1=0.9704 | val_f1=0.5863 | val_bal_acc=0.5790 [image_age_location_encoded32] epoch 17/20 | train_f1=0.9770 | val_f1=0.5887 | val_bal_acc=0.5706 [image_age_location_encoded32] epoch 18/20 | train_f1=0.9746 | val_f1=0.5928 | val_bal_acc=0.5809 [image_age_location_encoded32] epoch 19/20 | train_f1=0.9831 | val_f1=0.5890 | val_bal_acc=0.5970 [image_age_location_encoded32] epoch 20/20 | train_f1=0.9847 | val_f1=0.5976 | val_bal_acc=0.5815 ===== Running image_age_location_encoded64 ===== [image_age_location_encoded64] epoch 1/20 | train_f1=0.4855 | val_f1=0.5385 | val_bal_acc=0.5974 [image_age_location_encoded64] epoch 2/20 | train_f1=0.7095 | val_f1=0.5459 | val_bal_acc=0.6065 [image_age_location_encoded64] epoch 3/20 | train_f1=0.7730 | val_f1=0.5723 | val_bal_acc=0.5852 [image_age_location_encoded64] epoch 4/20 | train_f1=0.8275 | val_f1=0.5843 | val_bal_acc=0.6092 [image_age_location_encoded64] epoch 5/20 | train_f1=0.8622 | val_f1=0.5560 | val_bal_acc=0.5951 [image_age_location_encoded64] epoch 6/20 | train_f1=0.8896 | val_f1=0.5924 | val_bal_acc=0.5971 [image_age_location_encoded64] epoch 7/20 | train_f1=0.9107 | val_f1=0.5917 | val_bal_acc=0.6013 [image_age_location_encoded64] epoch 8/20 | train_f1=0.9143 | val_f1=0.5972 | val_bal_acc=0.6078 [image_age_location_encoded64] epoch 9/20 | train_f1=0.9243 | val_f1=0.5710 | val_bal_acc=0.5926 [image_age_location_encoded64] epoch 10/20 | train_f1=0.9357 | val_f1=0.5849 | val_bal_acc=0.5874 [image_age_location_encoded64] epoch 11/20 | train_f1=0.9477 | val_f1=0.5990 | val_bal_acc=0.5803 [image_age_location_encoded64] epoch 12/20 | train_f1=0.9526 | val_f1=0.5753 | val_bal_acc=0.5658 [image_age_location_encoded64] epoch 13/20 | train_f1=0.9605 | val_f1=0.5892 | val_bal_acc=0.5525 [image_age_location_encoded64] epoch 14/20 | train_f1=0.9631 | val_f1=0.5648 | val_bal_acc=0.5769 [image_age_location_encoded64] epoch 15/20 | train_f1=0.9668 | val_f1=0.5976 | val_bal_acc=0.5747 [image_age_location_encoded64] epoch 16/20 | train_f1=0.9724 | val_f1=0.5973 | val_bal_acc=0.5725 [image_age_location_encoded64] epoch 17/20 | train_f1=0.9743 | val_f1=0.5868 | val_bal_acc=0.5863 [image_age_location_encoded64] epoch 18/20 | train_f1=0.9794 | val_f1=0.5905 | val_bal_acc=0.5746 [image_age_location_encoded64] epoch 19/20 | train_f1=0.9832 | val_f1=0.5954 | val_bal_acc=0.5735 [image_age_location_encoded64] epoch 20/20 | train_f1=0.9860 | val_f1=0.5959 | val_bal_acc=0.5616 ===== Running image_age_location_encoded128 ===== [image_age_location_encoded128] epoch 1/20 | train_f1=0.4778 | val_f1=0.4868 | val_bal_acc=0.6182 [image_age_location_encoded128] epoch 2/20 | train_f1=0.6964 | val_f1=0.5242 | val_bal_acc=0.6132 [image_age_location_encoded128] epoch 3/20 | train_f1=0.7921 | val_f1=0.5660 | val_bal_acc=0.6268 [image_age_location_encoded128] epoch 4/20 | train_f1=0.8167 | val_f1=0.5304 | val_bal_acc=0.5859 [image_age_location_encoded128] epoch 5/20 | train_f1=0.8568 | val_f1=0.5606 | val_bal_acc=0.5898 [image_age_location_encoded128] epoch 6/20 | train_f1=0.8787 | val_f1=0.5698 | val_bal_acc=0.5943 [image_age_location_encoded128] epoch 7/20 | train_f1=0.8998 | val_f1=0.5775 | val_bal_acc=0.6002 [image_age_location_encoded128] epoch 8/20 | train_f1=0.9151 | val_f1=0.5624 | val_bal_acc=0.5806 [image_age_location_encoded128] epoch 9/20 | train_f1=0.9249 | val_f1=0.6040 | val_bal_acc=0.5864 [image_age_location_encoded128] epoch 10/20 | train_f1=0.9334 | val_f1=0.5929 | val_bal_acc=0.5830 [image_age_location_encoded128] epoch 11/20 | train_f1=0.9412 | val_f1=0.5795 | val_bal_acc=0.5959 [image_age_location_encoded128] epoch 12/20 | train_f1=0.9452 | val_f1=0.5883 | val_bal_acc=0.5926 [image_age_location_encoded128] epoch 13/20 | train_f1=0.9541 | val_f1=0.5706 | val_bal_acc=0.5817 [image_age_location_encoded128] epoch 14/20 | train_f1=0.9648 | val_f1=0.5996 | val_bal_acc=0.5824 [image_age_location_encoded128] epoch 15/20 | train_f1=0.9673 | val_f1=0.6093 | val_bal_acc=0.5984 [image_age_location_encoded128] epoch 16/20 | train_f1=0.9732 | val_f1=0.5980 | val_bal_acc=0.5895 [image_age_location_encoded128] epoch 17/20 | train_f1=0.9769 | val_f1=0.5956 | val_bal_acc=0.5815 [image_age_location_encoded128] epoch 18/20 | train_f1=0.9757 | val_f1=0.5915 | val_bal_acc=0.5667 [image_age_location_encoded128] epoch 19/20 | train_f1=0.9804 | val_f1=0.6062 | val_bal_acc=0.5890 [image_age_location_encoded128] epoch 20/20 | train_f1=0.9810 | val_f1=0.6023 | val_bal_acc=0.5837 ===== Running image_sex_location_encoded32 ===== [image_sex_location_encoded32] epoch 1/20 | train_f1=0.4890 | val_f1=0.5574 | val_bal_acc=0.6146 [image_sex_location_encoded32] epoch 2/20 | train_f1=0.7173 | val_f1=0.5409 | val_bal_acc=0.6004 [image_sex_location_encoded32] epoch 3/20 | train_f1=0.8031 | val_f1=0.5374 | val_bal_acc=0.5791 [image_sex_location_encoded32] epoch 4/20 | train_f1=0.8265 | val_f1=0.5526 | val_bal_acc=0.6012 [image_sex_location_encoded32] epoch 5/20 | train_f1=0.8504 | val_f1=0.5729 | val_bal_acc=0.5927 [image_sex_location_encoded32] epoch 6/20 | train_f1=0.8754 | val_f1=0.6057 | val_bal_acc=0.5885 [image_sex_location_encoded32] epoch 7/20 | train_f1=0.8973 | val_f1=0.5808 | val_bal_acc=0.6099 [image_sex_location_encoded32] epoch 8/20 | train_f1=0.9118 | val_f1=0.5928 | val_bal_acc=0.5912 [image_sex_location_encoded32] epoch 9/20 | train_f1=0.9312 | val_f1=0.5862 | val_bal_acc=0.5921 [image_sex_location_encoded32] epoch 10/20 | train_f1=0.9294 | val_f1=0.5951 | val_bal_acc=0.5940 [image_sex_location_encoded32] epoch 11/20 | train_f1=0.9438 | val_f1=0.5729 | val_bal_acc=0.6025 [image_sex_location_encoded32] epoch 12/20 | train_f1=0.9520 | val_f1=0.5895 | val_bal_acc=0.6009 [image_sex_location_encoded32] epoch 13/20 | train_f1=0.9503 | val_f1=0.5883 | val_bal_acc=0.5781 [image_sex_location_encoded32] epoch 14/20 | train_f1=0.9624 | val_f1=0.6013 | val_bal_acc=0.5705 [image_sex_location_encoded32] epoch 15/20 | train_f1=0.9688 | val_f1=0.5870 | val_bal_acc=0.5686 [image_sex_location_encoded32] epoch 16/20 | train_f1=0.9691 | val_f1=0.5898 | val_bal_acc=0.5495 [image_sex_location_encoded32] epoch 17/20 | train_f1=0.9756 | val_f1=0.5960 | val_bal_acc=0.5883 [image_sex_location_encoded32] epoch 18/20 | train_f1=0.9780 | val_f1=0.5769 | val_bal_acc=0.5847 [image_sex_location_encoded32] epoch 19/20 | train_f1=0.9752 | val_f1=0.5991 | val_bal_acc=0.5799 [image_sex_location_encoded32] epoch 20/20 | train_f1=0.9816 | val_f1=0.5980 | val_bal_acc=0.5595 ===== Running image_sex_location_encoded64 ===== [image_sex_location_encoded64] epoch 1/20 | train_f1=0.4979 | val_f1=0.5109 | val_bal_acc=0.5874 [image_sex_location_encoded64] epoch 2/20 | train_f1=0.7054 | val_f1=0.5586 | val_bal_acc=0.6199 [image_sex_location_encoded64] epoch 3/20 | train_f1=0.7800 | val_f1=0.5460 | val_bal_acc=0.6050 [image_sex_location_encoded64] epoch 4/20 | train_f1=0.8249 | val_f1=0.5652 | val_bal_acc=0.6114 [image_sex_location_encoded64] epoch 5/20 | train_f1=0.8469 | val_f1=0.5552 | val_bal_acc=0.5900 [image_sex_location_encoded64] epoch 6/20 | train_f1=0.8782 | val_f1=0.5904 | val_bal_acc=0.6008 [image_sex_location_encoded64] epoch 7/20 | train_f1=0.8895 | val_f1=0.5937 | val_bal_acc=0.5821 [image_sex_location_encoded64] epoch 8/20 | train_f1=0.9091 | val_f1=0.5854 | val_bal_acc=0.5953 [image_sex_location_encoded64] epoch 9/20 | train_f1=0.9196 | val_f1=0.5754 | val_bal_acc=0.5810 [image_sex_location_encoded64] epoch 10/20 | train_f1=0.9319 | val_f1=0.5824 | val_bal_acc=0.5950 [image_sex_location_encoded64] epoch 11/20 | train_f1=0.9409 | val_f1=0.5950 | val_bal_acc=0.5842 [image_sex_location_encoded64] epoch 12/20 | train_f1=0.9514 | val_f1=0.5966 | val_bal_acc=0.5899 [image_sex_location_encoded64] epoch 13/20 | train_f1=0.9587 | val_f1=0.5830 | val_bal_acc=0.5867 [image_sex_location_encoded64] epoch 14/20 | train_f1=0.9622 | val_f1=0.6011 | val_bal_acc=0.5811 [image_sex_location_encoded64] epoch 15/20 | train_f1=0.9618 | val_f1=0.5883 | val_bal_acc=0.5646 [image_sex_location_encoded64] epoch 16/20 | train_f1=0.9735 | val_f1=0.5813 | val_bal_acc=0.5815 [image_sex_location_encoded64] epoch 17/20 | train_f1=0.9713 | val_f1=0.6125 | val_bal_acc=0.5907 [image_sex_location_encoded64] epoch 18/20 | train_f1=0.9814 | val_f1=0.5778 | val_bal_acc=0.5980 [image_sex_location_encoded64] epoch 19/20 | train_f1=0.9779 | val_f1=0.5768 | val_bal_acc=0.5793 [image_sex_location_encoded64] epoch 20/20 | train_f1=0.9859 | val_f1=0.6074 | val_bal_acc=0.5725 ===== Running image_sex_location_encoded128 ===== [image_sex_location_encoded128] epoch 1/20 | train_f1=0.4777 | val_f1=0.4964 | val_bal_acc=0.6232 [image_sex_location_encoded128] epoch 2/20 | train_f1=0.6931 | val_f1=0.5550 | val_bal_acc=0.6458 [image_sex_location_encoded128] epoch 3/20 | train_f1=0.7765 | val_f1=0.5703 | val_bal_acc=0.6463 [image_sex_location_encoded128] epoch 4/20 | train_f1=0.8177 | val_f1=0.5527 | val_bal_acc=0.6006 [image_sex_location_encoded128] epoch 5/20 | train_f1=0.8547 | val_f1=0.5748 | val_bal_acc=0.5960 [image_sex_location_encoded128] epoch 6/20 | train_f1=0.8717 | val_f1=0.5563 | val_bal_acc=0.5875 [image_sex_location_encoded128] epoch 7/20 | train_f1=0.8887 | val_f1=0.5859 | val_bal_acc=0.5815 [image_sex_location_encoded128] epoch 8/20 | train_f1=0.9063 | val_f1=0.5678 | val_bal_acc=0.5716 [image_sex_location_encoded128] epoch 9/20 | train_f1=0.9207 | val_f1=0.5816 | val_bal_acc=0.5830 [image_sex_location_encoded128] epoch 10/20 | train_f1=0.9300 | val_f1=0.6055 | val_bal_acc=0.5963 [image_sex_location_encoded128] epoch 11/20 | train_f1=0.9372 | val_f1=0.5656 | val_bal_acc=0.5860 [image_sex_location_encoded128] epoch 12/20 | train_f1=0.9411 | val_f1=0.5852 | val_bal_acc=0.5881 [image_sex_location_encoded128] epoch 13/20 | train_f1=0.9526 | val_f1=0.5943 | val_bal_acc=0.5652 [image_sex_location_encoded128] epoch 14/20 | train_f1=0.9635 | val_f1=0.6032 | val_bal_acc=0.5789 [image_sex_location_encoded128] epoch 15/20 | train_f1=0.9651 | val_f1=0.6066 | val_bal_acc=0.5754 [image_sex_location_encoded128] epoch 16/20 | train_f1=0.9675 | val_f1=0.5946 | val_bal_acc=0.5661 [image_sex_location_encoded128] epoch 17/20 | train_f1=0.9729 | val_f1=0.5890 | val_bal_acc=0.5803 [image_sex_location_encoded128] epoch 18/20 | train_f1=0.9757 | val_f1=0.5989 | val_bal_acc=0.5765 [image_sex_location_encoded128] epoch 19/20 | train_f1=0.9755 | val_f1=0.5978 | val_bal_acc=0.5770 [image_sex_location_encoded128] epoch 20/20 | train_f1=0.9750 | val_f1=0.5865 | val_bal_acc=0.5556
Out[25]:
18
In [26]:
# Cell C: 整理成总表，并挑每个组合的最佳维度
import pandas as pd
all_search_df = pd.DataFrame(all_search_results)
display(
all_search_df[
[
"method",
"base_method",
"features",
"metadata_dim",
"metadata_embed_dim",
"best_val_macro_f1",
"test_accuracy",
"test_balanced_accuracy",
"test_macro_f1",
]
].sort_values(["base_method", "metadata_embed_dim"])
)
best_per_combination_df = (
all_search_df
.sort_values(["base_method", "best_val_macro_f1"], ascending=[True, False])
.groupby("base_method", as_index=False)
.first()
)
display(
best_per_combination_df[
[
"method",
"base_method",
"features",
"metadata_dim",
"metadata_embed_dim",
"best_val_macro_f1",
"test_accuracy",
"test_balanced_accuracy",
"test_macro_f1",
]
].sort_values("base_method")
)
| method | base_method | features | metadata_dim | metadata_embed_dim | best_val_macro_f1 | test_accuracy | test_balanced_accuracy | test_macro_f1 | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | image_age_encoded32 | image_age | age | 1 | 32 | 0.592534 | 0.770425 | 0.578116 | 0.576380 |
| 1 | image_age_encoded64 | image_age | age | 1 | 64 | 0.593023 | 0.781904 | 0.601570 | 0.584162 |
| 2 | image_age_encoded128 | image_age | age | 1 | 128 | 0.594447 | 0.766374 | 0.549634 | 0.557388 |
| 12 | image_age_location_encoded32 | image_age_location | age,location | 16 | 32 | 0.604529 | 0.799460 | 0.578341 | 0.588440 |
| 13 | image_age_location_encoded64 | image_age_location | age,location | 16 | 64 | 0.599031 | 0.788656 | 0.568477 | 0.575066 |
| 14 | image_age_location_encoded128 | image_age_location | age,location | 16 | 128 | 0.609298 | 0.783255 | 0.602824 | 0.607910 |
| 9 | image_age_sex_encoded32 | image_age_sex | age,sex | 4 | 32 | 0.603107 | 0.777178 | 0.587462 | 0.590630 |
| 10 | image_age_sex_encoded64 | image_age_sex | age,sex | 4 | 64 | 0.604039 | 0.793383 | 0.582281 | 0.581666 |
| 11 | image_age_sex_encoded128 | image_age_sex | age,sex | 4 | 128 | 0.602657 | 0.774477 | 0.572578 | 0.576931 |
| 6 | image_location_encoded32 | image_location | location | 15 | 32 | 0.603778 | 0.754220 | 0.558853 | 0.564874 |
| 7 | image_location_encoded64 | image_location | location | 15 | 64 | 0.600608 | 0.781904 | 0.574504 | 0.580142 |
| 8 | image_location_encoded128 | image_location | location | 15 | 128 | 0.595590 | 0.784605 | 0.543721 | 0.561448 |
| 3 | image_sex_encoded32 | image_sex | sex | 3 | 32 | 0.596308 | 0.759622 | 0.575438 | 0.576118 |
| 4 | image_sex_encoded64 | image_sex | sex | 3 | 64 | 0.600935 | 0.786631 | 0.580348 | 0.579055 |
| 5 | image_sex_encoded128 | image_sex | sex | 3 | 128 | 0.599878 | 0.766374 | 0.568703 | 0.569694 |
| 15 | image_sex_location_encoded32 | image_sex_location | sex,location | 18 | 32 | 0.605750 | 0.791357 | 0.611511 | 0.615742 |
| 16 | image_sex_location_encoded64 | image_sex_location | sex,location | 18 | 64 | 0.612467 | 0.792708 | 0.598458 | 0.596497 |
| 17 | image_sex_location_encoded128 | image_sex_location | sex,location | 18 | 128 | 0.606623 | 0.796759 | 0.585301 | 0.596229 |
| method | base_method | features | metadata_dim | metadata_embed_dim | best_val_macro_f1 | test_accuracy | test_balanced_accuracy | test_macro_f1 | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | image_age_encoded128 | image_age | age | 1 | 128 | 0.594447 | 0.766374 | 0.549634 | 0.557388 |
| 1 | image_age_location_encoded128 | image_age_location | age,location | 16 | 128 | 0.609298 | 0.783255 | 0.602824 | 0.607910 |
| 2 | image_age_sex_encoded64 | image_age_sex | age,sex | 4 | 64 | 0.604039 | 0.793383 | 0.582281 | 0.581666 |
| 3 | image_location_encoded32 | image_location | location | 15 | 32 | 0.603778 | 0.754220 | 0.558853 | 0.564874 |
| 4 | image_sex_encoded64 | image_sex | sex | 3 | 64 | 0.600935 | 0.786631 | 0.580348 | 0.579055 |
| 5 | image_sex_location_encoded64 | image_sex_location | sex,location | 18 | 64 | 0.612467 | 0.792708 | 0.598458 | 0.596497 |
In [27]:
# Cell D: 保存结果
from datetime import datetime
import json
from pathlib import Path
save_dir = Path("/Users/applesues01/Documents/Medical_Agent/supports")
save_dir.mkdir(parents=True, exist_ok=True)
date_tag = datetime.now().strftime("%Y-%m-%d")
time_tag = datetime.now().strftime("%H%M%S")
full_csv = save_dir / f"{date_tag}_{time_tag}_metadata_embed_search_full.csv"
best_csv = save_dir / f"{date_tag}_{time_tag}_metadata_embed_search_best.csv"
summary_json = save_dir / f"{date_tag}_{time_tag}_metadata_embed_search_summary.json"
all_search_df.to_csv(full_csv, index=False)
best_per_combination_df.to_csv(best_csv, index=False)
payload = {
"generated_at": datetime.now().isoformat(),
"purpose": "best metadata embedding dimension search for 6 metadata combinations",
"embed_dims": EMBED_DIMS,
"num_epochs": 5,
"full_results": all_search_results,
"best_results": best_per_combination_df.to_dict(orient="records"),
}
def to_jsonable(x):
import numpy as np
if isinstance(x, dict):
return {k: to_jsonable(v) for k, v in x.items()}
if isinstance(x, list):
return [to_jsonable(v) for v in x]
if isinstance(x, tuple):
return [to_jsonable(v) for v in x]
if isinstance(x, np.integer):
return int(x)
if isinstance(x, np.floating):
return float(x)
if isinstance(x, np.ndarray):
return x.tolist()
return x
with open(summary_json, "w", encoding="utf-8") as f:
json.dump(to_jsonable(payload), f, ensure_ascii=False, indent=2)
print("Saved:")
print(full_csv)
print(best_csv)
print(summary_json)
Saved: /Users/applesues01/Documents/Medical_Agent/supports/2026-08-03_165248_metadata_embed_search_full.csv /Users/applesues01/Documents/Medical_Agent/supports/2026-08-03_165248_metadata_embed_search_best.csv /Users/applesues01/Documents/Medical_Agent/supports/2026-08-03_165248_metadata_embed_search_summary.json