"""Output helpers for the analysis tool layer."""
from __future__ import annotations
import base64
import html as html_module
import mimetypes
import shutil
from dataclasses import dataclass
from hashlib import sha1
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
try:
import pandas as pd
except ModuleNotFoundError: # pragma: no cover - optional dependency
pd = None
from agentsociety2.logger import get_logger
from .chart_export import (
EDA_HUB_ENTRIES,
brand_header_html,
eda_datatable_css,
eda_hub_page_css,
eda_index_page_css,
ensure_brand_icon,
export_plotly_html,
html_font_links,
html_tab_switcher_script,
)
from .models import (
DIR_ARTIFACTS,
DIR_EXPERIMENT_PREFIX,
DIR_HYPOTHESIS_PREFIX,
DIR_REPORT_ASSETS,
DIR_RUN,
SUPPORTED_IMAGE_FORMATS,
ReportAsset,
)
from .utils import _sanitize_id
logger = get_logger()
[文档]
@dataclass
class ReportPaths:
"""Report file paths."""
markdown: Path
html: Path
markdown_zh: Optional[Path] = None
html_zh: Optional[Path] = None
markdown_en: Optional[Path] = None
html_en: Optional[Path] = None
assets_dir: Optional[Path] = None
[文档]
class AssetManager:
"""Discover and normalize analysis assets."""
[文档]
def __init__(self, workspace_path: Path):
self.workspace_path = Path(workspace_path)
self.logger = logger
[文档]
def discover_assets(
self,
experiment_id: str,
hypothesis_id: str,
) -> List[ReportAsset]:
"""Discover image assets under `run/artifacts`."""
hid = _sanitize_id(hypothesis_id)
eid = _sanitize_id(experiment_id)
asset_path = (
self.workspace_path
/ f"{DIR_HYPOTHESIS_PREFIX}{hid}"
/ f"{DIR_EXPERIMENT_PREFIX}{eid}"
/ DIR_RUN
/ DIR_ARTIFACTS
)
assets: List[ReportAsset] = []
if not asset_path.exists():
return assets
for file_path in asset_path.rglob("*"):
if file_path.suffix.lower() not in SUPPORTED_IMAGE_FORMATS:
continue
relative_path = file_path.relative_to(asset_path)
assets.append(
ReportAsset(
asset_id=self._build_asset_id("viz", relative_path),
asset_type="visualization",
title=self._format_title(file_path.stem),
file_path=str(file_path),
description=f"Generated visualization: {relative_path.as_posix()}",
file_size=file_path.stat().st_size,
)
)
return assets
[文档]
def process_assets(
self,
assets: List[ReportAsset],
output_dir: Path,
*,
include_embedded_data: bool = True,
) -> Dict[str, Any]:
"""Copy assets into `assets/`; optionally generate base64 payloads."""
assets_dir = output_dir / DIR_REPORT_ASSETS
assets_dir.mkdir(exist_ok=True)
ensure_brand_icon(assets_dir)
processed: Dict[str, Any] = {}
used_names: set[str] = set()
for asset in assets:
source_path = Path(asset.file_path)
if not source_path.exists():
continue
dest_name = self._build_unique_dest_name(
source_path, asset.asset_id, used_names
)
dest_path = assets_dir / dest_name
if source_path.resolve() != dest_path.resolve():
shutil.copy2(source_path, dest_path)
embedded = None
if (
include_embedded_data
and source_path.suffix.lower() in SUPPORTED_IMAGE_FORMATS
):
with open(source_path, "rb") as file_handle:
encoded = base64.b64encode(file_handle.read()).decode("utf-8")
mime = (
mimetypes.guess_type(source_path.name) or ("image/png", None)
)[0]
embedded = f"data:{mime};base64,{encoded}"
processed[asset.asset_id] = {
"title": asset.title,
"local_path": str(dest_path),
"relative_path": f"{DIR_REPORT_ASSETS}/{dest_name}",
"embedded_data": embedded,
"description": asset.description,
}
return processed
def _build_asset_id(self, prefix: str, relative_path: Path) -> str:
normalized = relative_path.as_posix()
digest = sha1(normalized.encode("utf-8")).hexdigest()[:10]
return f"{prefix}_{relative_path.stem}_{digest}"
def _build_unique_dest_name(
self,
source_path: Path,
asset_id: str,
used_names: set[str],
) -> str:
candidate = source_path.name
if candidate not in used_names:
used_names.add(candidate)
return candidate
suffix = source_path.suffix
stem = source_path.stem
unique_name = f"{stem}_{asset_id[-10:]}{suffix}"
counter = 1
while unique_name in used_names:
unique_name = f"{stem}_{asset_id[-10:]}_{counter}{suffix}"
counter += 1
used_names.add(unique_name)
return unique_name
def _format_title(self, filename: str) -> str:
title = filename.replace("_", " ").replace("-", " ")
return " ".join(word.capitalize() for word in title.split())
AssetProcessor = AssetManager
[文档]
class EDAGenerator:
"""Generate exploratory data analysis artifacts."""
[文档]
def __init__(self):
self.logger = logger
[文档]
def resolve_table_selection(
self,
reader,
tables: Optional[List[str]],
) -> Tuple[List[str], List[str], List[str]]:
"""Return requested, selected, and invalid table names."""
available_tables = reader.read_schema().tables
if tables is None:
return available_tables, available_tables, []
requested_tables: List[str] = []
seen = set()
for table in tables:
name = (table or "").strip()
if not name or name in seen:
continue
seen.add(name)
requested_tables.append(name)
available_set = set(available_tables)
selected_tables = [
table for table in requested_tables if table in available_set
]
invalid_tables = [
table for table in requested_tables if table not in available_set
]
return requested_tables, selected_tables, invalid_tables
def _resolve_tables(self, reader, tables: Optional[List[str]]) -> List[str]:
_, selected_tables, _ = self.resolve_table_selection(reader, tables)
return selected_tables
[文档]
def generate_quick_stats(
self,
db_path: Path,
max_rows: int = 5000,
tables: Optional[List[str]] = None,
) -> Optional[str]:
if not db_path.exists():
return None
from .data import DataReader, DatabaseSchema
reader = DataReader(db_path)
schema = reader.read_schema()
selected_tables = self._resolve_tables(reader, tables)
filtered_schema = DatabaseSchema(
tables=selected_tables,
columns={table: schema.columns.get(table, []) for table in selected_tables},
row_counts={
table: schema.row_counts.get(table, 0) for table in selected_tables
},
markdown=schema.markdown,
)
stats = reader.compute_stats(filtered_schema)
return stats.quick_stats_md
[文档]
def generate_missingno_report(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 50000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
"""Generate a missing-value report with missingno."""
if not db_path.exists():
return None
from .data import DataReader
reader = DataReader(db_path)
selected_tables = self._resolve_tables(reader, tables)
sample = reader.read_sample_data(tables=selected_tables, limit=max_rows)
if not sample:
return None
if pd is None:
self.logger.warning("pandas 未安装，跳过 missingno 报告生成")
return None
all_dfs = []
for table_name, data in sample.items():
if data and len(data) > 0:
df = pd.DataFrame(data)
df.columns = [f"{table_name}.{col}" for col in df.columns]
all_dfs.append(df)
if not all_dfs:
return None
combined_df = pd.concat(all_dfs, axis=1)
if len(combined_df.columns) > 50:
missing_counts = combined_df.isnull().sum()
top_missing = missing_counts.nlargest(50).index.tolist()
combined_df = combined_df[top_missing]
output_dir.mkdir(parents=True, exist_ok=True)
try:
import matplotlib.pyplot as plt
import missingno as msno
out_file = output_dir / "eda_missingno.html"
_fig, axes = plt.subplots(2, 2, figsize=(16, 12))
try:
msno.matrix(combined_df, ax=axes[0, 0], fontsize=8)
axes[0, 0].set_title("Missing Value Matrix", fontsize=12)
except Exception:
axes[0, 0].text(
0.5, 0.5, "Matrix plot failed", ha="center", va="center"
)
axes[0, 0].set_title("Missing Value Matrix (Error)")
try:
msno.bar(combined_df, ax=axes[0, 1], fontsize=8)
axes[0, 1].set_title("Missing Value Bar Chart", fontsize=12)
except Exception:
axes[0, 1].text(0.5, 0.5, "Bar plot failed", ha="center", va="center")
axes[0, 1].set_title("Missing Value Bar (Error)")
try:
if len(combined_df.columns) > 1:
msno.heatmap(combined_df, ax=axes[1, 0], fontsize=8)
axes[1, 0].set_title(
"Missing Value Correlation Heatmap", fontsize=12
)
else:
axes[1, 0].text(
0.5, 0.5, "Need >1 columns", ha="center", va="center"
)
axes[1, 0].set_title("Correlation Heatmap (Skipped)")
except Exception:
axes[1, 0].text(0.5, 0.5, "Heatmap failed", ha="center", va="center")
axes[1, 0].set_title("Correlation Heatmap (Error)")
try:
if len(combined_df.columns) > 1:
msno.dendrogram(combined_df, ax=axes[1, 1], fontsize=8)
axes[1, 1].set_title("Missing Value Dendrogram", fontsize=12)
else:
axes[1, 1].text(
0.5, 0.5, "Need >1 columns", ha="center", va="center"
)
axes[1, 1].set_title("Dendrogram (Skipped)")
except Exception:
axes[1, 1].text(0.5, 0.5, "Dendrogram failed", ha="center", va="center")
axes[1, 1].set_title("Dendrogram (Error)")
plt.tight_layout()
plt.savefig(
str(out_file).replace(".html", ".png"), dpi=150, bbox_inches="tight"
)
plt.close()
html_content = f"""<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Missing Value Analysis (missingno)</title>
<style>
body {{ font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; margin: 40px; }}
h1 {{ color: #333; }}
.summary {{ background: #f5f5f5; padding: 15px; border-radius: 5px; margin: 20px 0; }}
.stats {{ margin-top: 10px; }}
img {{ max-width: 100%; height: auto; margin: 20px 0; border: 1px solid #ddd; }}
</style>
</head>
<body>
<h1>Missing Value Analysis</h1>
<div class="summary">
<p><strong>Tool</strong>: missingno - Missing data visualization</p>
<p><strong>Total Columns Analyzed</strong>: {len(combined_df.columns)}</p>
<p><strong>Total Rows</strong>: {len(combined_df)}</p>
<div class="stats">
<p><strong>Total Missing Values</strong>: {combined_df.isnull().sum().sum()}</p>
<p><strong>Columns with Missing</strong>: {(combined_df.isnull().sum() > 0).sum()}</p>
</div>
</div>
<img src="eda_missingno.png" alt="Missing Value Visualization">
</body>
</html>"""
out_file.write_text(html_content, encoding="utf-8")
self.logger.info("生成 missingno 缺失值报告: %s", out_file)
return out_file
except ImportError:
self.logger.warning("missingno 未安装，跳过缺失值可视化")
return None
except Exception as exc:
self.logger.warning("missingno 生成失败: %s", exc)
return None
[文档]
def generate_correlation_report(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 50000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
"""Generate a correlation report."""
if not db_path.exists():
return None
from .data import DataReader
reader = DataReader(db_path)
selected_tables = self._resolve_tables(reader, tables)
sample = reader.read_sample_data(tables=selected_tables, limit=max_rows)
if not sample:
return None
if pd is None:
self.logger.warning("pandas 未安装，跳过相关性报告生成")
return None
output_dir.mkdir(parents=True, exist_ok=True)
generated_files = []
for table_name, data in sample.items():
if not data or len(data) < 2:
continue
df = pd.DataFrame(data)
numeric_df = df.select_dtypes(include=["number"])
if len(numeric_df.columns) < 2:
continue
try:
import matplotlib.pyplot as plt
import seaborn as sns
corr_matrix = numeric_df.corr()
_fig, ax = plt.subplots(figsize=(12, 10))
sns.heatmap(
corr_matrix,
annot=True,
fmt=".2f",
cmap="RdBu_r",
center=0,
square=True,
linewidths=0.5,
ax=ax,
)
ax.set_title(f"Correlation Matrix: {table_name}", fontsize=14)
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
out_file = output_dir / f"correlation_{safe_name}.png"
plt.tight_layout()
plt.savefig(str(out_file), dpi=150, bbox_inches="tight")
plt.close()
generated_files.append((table_name, str(out_file)))
self.logger.info(
"生成相关性矩阵: %s (表: %s, %d 列)",
out_file,
table_name,
len(numeric_df.columns),
)
except Exception as exc:
self.logger.warning("生成表 %s 的相关性矩阵失败: %s", table_name, exc)
if not generated_files:
return None
index_file = output_dir / "correlation_index.html"
rows = "\n".join(
f'<tr><td>{name}</td><td><img src="{Path(path).name}" style="max-width:800px"></td></tr>'
for name, path in generated_files
)
html_content = f"""<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Correlation Analysis</title>
<style>
body {{ font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; margin: 40px; }}
h1 {{ color: #333; }}
table {{ border-collapse: collapse; width: 100%; }}
td {{ padding: 20px; border-bottom: 1px solid #ddd; }}
img {{ max-width: 100%; }}
</style>
</head>
<body>
<h1>Correlation Analysis Report</h1>
<table>
<tr><th>Table</th><th>Correlation Matrix</th></tr>
{rows}
</table>
</body>
</html>"""
index_file.write_text(html_content, encoding="utf-8")
return index_file
[文档]
def generate_ydata_profile(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 10000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
"""Generate a ydata-profiling report."""
if not db_path.exists():
return None
from .data import DataReader
reader = DataReader(db_path)
selected_tables = self._resolve_tables(reader, tables)
sample = reader.read_sample_data(tables=selected_tables, limit=max_rows)
if not sample:
return None
if pd is None:
self.logger.warning("pandas 未安装，跳过 ydata-profiling 报告生成")
return None
non_empty_tables = {
table_name: data
for table_name, data in sample.items()
if data and len(data) > 0
}
if not non_empty_tables:
return None
output_dir.mkdir(parents=True, exist_ok=True)
try:
from ydata_profiling import ProfileReport
if len(non_empty_tables) == 1:
table_name = next(iter(non_empty_tables.keys()))
df = pd.DataFrame(non_empty_tables[table_name])
out_file = output_dir / "eda_profile.html"
profile = ProfileReport(df, title=f"EDA: {table_name}", minimal=True)
profile.to_file(str(out_file))
self.logger.info(
"生成 ydata-profiling 报告: %s (表: %s, %d 行)",
out_file,
table_name,
len(df),
)
return out_file
generated_files = []
for table_name, data in non_empty_tables.items():
df = pd.DataFrame(data)
if df.empty:
continue
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
table_file = output_dir / f"eda_profile_{safe_name}.html"
try:
profile = ProfileReport(
df, title=f"EDA: {table_name}", minimal=True
)
profile.to_file(str(table_file))
generated_files.append((table_name, table_file.name, len(df)))
self.logger.info(
"生成 ydata-profiling 表报告: %s (表: %s, %d 行)",
table_file,
table_name,
len(df),
)
except Exception as exc:
self.logger.warning(
"生成表 %s 的 EDA 报告失败: %s", table_name, exc
)
if not generated_files:
return None
index_file = output_dir / "eda_profile.html"
index_content = self._build_eda_index_html(
generated_files, "ydata-profiling"
)
index_file.write_text(index_content, encoding="utf-8")
self.logger.info(
"生成 EDA 索引页: %s (%d 张表)", index_file, len(generated_files)
)
return index_file
except Exception as exc:
self.logger.debug("ydata-profiling 生成失败: %s", exc)
return None
def _build_eda_index_html(
self,
table_files: List[Tuple[str, str, int]],
tool_name: str,
) -> str:
rows = "\n".join(
f'<tr><td><a href="{filename}">{name}</a></td><td class="num">{row_count}</td></tr>'
for name, filename, row_count in table_files
)
subtitle = f"{tool_name} · 选择表打开完整画像"
header = brand_header_html(
"交互式 EDA 索引",
subtitle,
variant="hub",
lang="zh",
)
return f"""<!DOCTYPE html>
<html lang="zh">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>EDA Index ({tool_name})</title>
{html_font_links()}
<style>{eda_index_page_css()}</style>
</head>
<body>
<div class="index-shell">
{header}
<div class="index-body">
<p class="panel-intro">分析报告可通过 <code>iframe</code> 嵌入下方各表画像。</p>
<table>
<tr><th>表</th><th class="num">行数</th></tr>
{rows}
</table>
</div>
</div>
</body>
</html>"""
[文档]
def generate_sweetviz_profile(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 10000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
"""Generate a Sweetviz report."""
if not db_path.exists():
return None
from .data import DataReader
reader = DataReader(db_path)
selected_tables = self._resolve_tables(reader, tables)
sample = reader.read_sample_data(tables=selected_tables, limit=max_rows)
if not sample:
return None
if pd is None:
self.logger.warning("pandas 未安装，跳过 Sweetviz 报告生成")
return None
non_empty_tables = {
table_name: data
for table_name, data in sample.items()
if data and len(data) > 0
}
if not non_empty_tables:
return None
output_dir.mkdir(parents=True, exist_ok=True)
try:
import sweetviz as sv
if len(non_empty_tables) == 1:
table_name = next(iter(non_empty_tables.keys()))
df = pd.DataFrame(non_empty_tables[table_name])
out_file = output_dir / "eda_sweetviz.html"
report = sv.analyze(df)
report.show_html(str(out_file), open_browser=False)
if out_file.exists():
self.logger.info(
"生成 Sweetviz 报告: %s (表: %s, %d 行)",
out_file,
table_name,
len(df),
)
return out_file
return None
generated_files = []
for table_name, data in non_empty_tables.items():
df = pd.DataFrame(data)
if df.empty:
continue
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
table_file = output_dir / f"eda_sweetviz_{safe_name}.html"
try:
report = sv.analyze(df)
report.show_html(str(table_file), open_browser=False)
if table_file.exists():
generated_files.append((table_name, table_file.name, len(df)))
self.logger.info(
"生成 Sweetviz 表报告: %s (表: %s, %d 行)",
table_file,
table_name,
len(df),
)
except Exception as exc:
self.logger.warning(
"生成表 %s 的 Sweetviz 报告失败: %s", table_name, exc
)
if not generated_files:
return None
index_file = output_dir / "eda_sweetviz.html"
index_content = self._build_eda_index_html(generated_files, "Sweetviz")
index_file.write_text(index_content, encoding="utf-8")
self.logger.info(
"生成 Sweetviz EDA 索引页: %s (%d 张表)",
index_file,
len(generated_files),
)
return index_file
except Exception as exc:
self.logger.debug("Sweetviz 生成失败: %s", exc)
return None
def _read_non_empty_tables(
self,
db_path: Path,
output_dir: Path,
max_rows: int,
tables: Optional[List[str]],
) -> Dict[str, "pd.DataFrame"]:
if pd is None or not db_path.exists():
return {}
from .data import DataReader
reader = DataReader(db_path)
selected_tables = self._resolve_tables(reader, tables)
sample = reader.read_sample_data(tables=selected_tables, limit=max_rows)
frames: Dict[str, pd.DataFrame] = {}
for table_name, data in (sample or {}).items():
if not data:
continue
df = pd.DataFrame(data)
if not df.empty:
frames[table_name] = df
return frames
[文档]
def generate_pygwalker_profile(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 10000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
frames = self._read_non_empty_tables(db_path, output_dir, max_rows, tables)
if not frames:
return None
output_dir.mkdir(parents=True, exist_ok=True)
try:
import pygwalker as pyg
except ImportError:
self.logger.warning("pygwalker 未安装，跳过拖拽探索报告")
return None
generated: List[Tuple[str, str, int]] = []
for table_name, df in frames.items():
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
out_file = output_dir / f"eda_pygwalker_{safe_name}.html"
html = pyg.to_html(df, embed_lib=True)
out_file.write_text(html, encoding="utf-8")
generated.append((table_name, out_file.name, len(df)))
self.logger.info("生成 PyGWalker 报告: %s", out_file)
if not generated:
return None
if len(generated) == 1:
single = output_dir / generated[0][1]
hub = output_dir / "eda_pygwalker.html"
if hub != single:
hub.write_text(single.read_text(encoding="utf-8"), encoding="utf-8")
return hub
index_file = output_dir / "eda_pygwalker.html"
index_file.write_text(
self._build_eda_index_html(generated, "PyGWalker"), encoding="utf-8"
)
return index_file
[文档]
def generate_datatable_profile(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 5000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
frames = self._read_non_empty_tables(db_path, output_dir, max_rows, tables)
if not frames:
return None
output_dir.mkdir(parents=True, exist_ok=True)
generated: List[Tuple[str, str, int]] = []
for table_name, df in frames.items():
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
out_file = output_dir / f"eda_datatable_{safe_name}.html"
out_file.write_text(
self._build_sortable_datatable_html(table_name, df),
encoding="utf-8",
)
generated.append((table_name, out_file.name, len(df)))
self.logger.info("生成可排序数据表: %s", out_file)
index_file = output_dir / "eda_datatable.html"
if len(generated) == 1:
index_file.write_text(
(output_dir / generated[0][1]).read_text(encoding="utf-8"),
encoding="utf-8",
)
else:
index_file.write_text(
self._build_eda_index_html(generated, "Interactive tables"),
encoding="utf-8",
)
return index_file
def _build_sortable_datatable_html(
self, table_name: str, df: "pd.DataFrame"
) -> str:
preview = df.head(500)
esc = html_module.escape
headers = "".join(f"<th>{esc(str(col))}</th>" for col in preview.columns)
rows = []
for _, row in preview.iterrows():
cells = "".join(f"<td>{esc(str(row[col]))}</td>" for col in preview.columns)
rows.append(f"<tr>{cells}</tr>")
body = "\n".join(rows)
truncated = len(df) > len(preview)
note = (
f"显示前 {len(preview)} / {len(df)} 行（完整数据见 sqlite）"
if truncated
else f"共 {len(df)} 行"
)
return f"""<!DOCTYPE html>
<html lang="zh">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Table: {esc(table_name)}</title>
{html_font_links()}
<style>{eda_datatable_css()}</style>
</head>
<body>
<div class="standalone-wrap">
{brand_header_html(esc(table_name), esc(note), variant="standalone", lang="zh")}
<div class="standalone-body">
<p class="meta">{note}</p>
<div class="toolbar">
<input type="search" id="filter" placeholder="筛选任意列…" />
</div>
<div class="wrap">
<table id="tbl">
<thead><tr>{headers}</tr></thead>
<tbody id="tbody">{body}</tbody>
</table>
</div>
<script>
(function () {{
var tbl = document.getElementById("tbl");
var filter = document.getElementById("filter");
var tbody = document.getElementById("tbody");
var rows = Array.prototype.slice.call(tbody.querySelectorAll("tr"));
filter.addEventListener("input", function () {{
var q = filter.value.toLowerCase();
rows.forEach(function (tr) {{
tr.style.display = tr.textContent.toLowerCase().indexOf(q) >= 0 ? "" : "none";
}});
}});
tbl.querySelectorAll("th").forEach(function (th, idx) {{
th.addEventListener("click", function () {{
var asc = th.dataset.asc !== "1";
tbl.querySelectorAll("th").forEach(function (h) {{ delete h.dataset.asc; }});
th.dataset.asc = asc ? "1" : "0";
rows.sort(function (a, b) {{
var av = (a.children[idx] && a.children[idx].textContent) || "";
var bv = (b.children[idx] && b.children[idx].textContent) || "";
var an = parseFloat(av), bn = parseFloat(bv);
if (!isNaN(an) && !isNaN(bn)) return asc ? an - bn : bn - an;
return asc ? av.localeCompare(bv) : bv.localeCompare(av);
}});
rows.forEach(function (tr) {{ tbody.appendChild(tr); }});
}});
}});
}})();
</script>
</div></div>
</body>
</html>"""
[文档]
def generate_plotly_profile(
self,
db_path: Path,
output_dir: Path,
max_rows: int = 8000,
tables: Optional[List[str]] = None,
) -> Optional[Path]:
frames = self._read_non_empty_tables(db_path, output_dir, max_rows, tables)
if not frames:
return None
output_dir.mkdir(parents=True, exist_ok=True)
try:
import plotly.express as px
except ImportError:
self.logger.warning("plotly 未安装，跳过 Plotly 概览")
return None
generated: List[Tuple[str, str, int]] = []
for table_name, df in frames.items():
numeric = df.select_dtypes(include=["number"])
if numeric.shape[1] < 1:
continue
cols = list(numeric.columns[:8])
safe_name = "".join(
c if c.isalnum() or c in "_-" else "_" for c in table_name
)
out_file = output_dir / f"eda_plotly_{safe_name}.html"
if len(cols) >= 2:
fig = px.scatter_matrix(
numeric[cols],
dimensions=cols,
title=f"Numeric overview: {table_name}",
)
else:
fig = px.histogram(
numeric,
x=cols[0],
title=f"Distribution: {table_name}.{cols[0]}",
)
export_plotly_html(fig, out_file)
generated.append((table_name, out_file.name, len(df)))
self.logger.info("生成 Plotly 概览: %s", out_file)
if not generated:
return None
index_file = output_dir / "eda_plotly.html"
index_file.write_text(
self._build_eda_index_html(generated, "Plotly"), encoding="utf-8"
)
return index_file
[文档]
def generate_eda_hub(
self,
output_dir: Path,
*,
title: str = "AgentSociety EDA Hub",
) -> Path:
output_dir.mkdir(parents=True, exist_ok=True)
assets_dir = output_dir.parent / "assets"
ensure_brand_icon(assets_dir)
entries = []
for spec in EDA_HUB_ENTRIES:
path = output_dir / spec["file"]
if path.is_file():
entries.append(spec)
hub_path = output_dir / "eda_hub.html"
hub_path.write_text(
self._build_eda_hub_html(entries, title=title, lang="zh"),
encoding="utf-8",
)
self.logger.info("生成 EDA Hub: %s (%d 入口)", hub_path, len(entries))
return hub_path
def _build_eda_hub_html(
self,
entries: List[Dict[str, str]],
*,
title: str,
lang: str = "zh",
) -> str:
zh = lang.startswith("zh")
subtitle = (
"多模式交互式数据探索 · 嵌入报告请使用 data/eda_hub.html"
if zh
else "Multi-mode interactive EDA · embed via data/eda_hub.html"
)
header = brand_header_html(title, subtitle, variant="hub", lang=lang)
if not entries:
body = '<p class="hub-empty">暂无 EDA 产物。请先运行 <code>run-eda --type bundle</code>。</p>'
else:
tab_buttons: List[str] = []
tab_panels: List[str] = []
for idx, spec in enumerate(entries):
active = " active" if idx == 0 else ""
tab_id = spec["key"]
label = spec["zh"] if zh else spec["en"]
desc = spec["desc_zh"] if zh else spec["desc_en"]
icon = spec["icon"]
tab_buttons.append(
f'<button type="button" class="tab{active}" data-tab="{tab_id}" '
f'role="tab" aria-selected="{"true" if idx == 0 else "false"}">'
f'<span class="tab-icon" aria-hidden="true">{icon}</span>'
f"<span>{label}</span></button>"
)
if spec["mode"] == "markdown":
md_path = spec["file"]
body_inner = (
f'<p class="panel-intro"><strong>{label}</strong> — {desc}</p>'
f'<p class="iframe-hint">'
f'<a href="{md_path}" target="_blank" rel="noopener">'
f'{"打开 Markdown" if zh else "Open Markdown"}</a></p>'
)
else:
open_lbl = "新标签页打开" if zh else "Open in new tab"
body_inner = (
f'<p class="panel-intro"><strong>{label}</strong> — {desc}</p>'
f'<p class="iframe-hint"><a href="{spec["file"]}" target="_blank" rel="noopener">{open_lbl}</a></p>'
f'<iframe class="eda-frame" src="{spec["file"]}" title="{label}" loading="lazy"></iframe>'
)
tab_panels.append(
f'<div class="tab-panel{active} {spec["panel"]}" data-panel="{tab_id}" role="tabpanel">'
f"{body_inner}</div>"
)
tabs_html = (
f'<div class="tab-root"><div class="tab-bar" role="tablist">{"".join(tab_buttons)}</div>'
f'{"".join(tab_panels)}</div>'
)
body = tabs_html
return f"""<!DOCTYPE html>
<html lang="{"zh" if zh else "en"}">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>{title}</title>
{html_font_links()}
<style>{eda_hub_page_css()}</style>
</head>
<body>
<div class="hub-shell">
{header}
<div class="hub-body">{body}</div>
</div>
{html_tab_switcher_script()}
</body>
</html>"""
[文档]
def generate_eda_bundle(
self,
db_path: Path,
output_dir: Path,
profiles: Optional[List[str]] = None,
tables: Optional[List[str]] = None,
max_rows: int = 10000,
) -> Tuple[List[str], Optional[Path]]:
default_profiles = [
"quick-stats",
"ydata",
"pygwalker",
"datatable",
"plotly-profile",
]
active = profiles or default_profiles
output_dir.mkdir(parents=True, exist_ok=True)
ensure_brand_icon(output_dir.parent / "assets")
files: List[str] = []
runners = {
"quick-stats": lambda: self._bundle_quick_stats(
db_path, output_dir, tables
),
"ydata": lambda: self.generate_ydata_profile(
db_path, output_dir, max_rows=max_rows, tables=tables
),
"sweetviz": lambda: self.generate_sweetviz_profile(
db_path, output_dir, max_rows=max_rows, tables=tables
),
"missingno": lambda: self.generate_missingno_report(
db_path, output_dir, tables=tables
),
"correlation": lambda: self.generate_correlation_report(
db_path, output_dir, tables=tables
),
"pygwalker": lambda: self.generate_pygwalker_profile(
db_path, output_dir, max_rows=max_rows, tables=tables
),
"datatable": lambda: self.generate_datatable_profile(
db_path, output_dir, max_rows=max_rows, tables=tables
),
"plotly-profile": lambda: self.generate_plotly_profile(
db_path, output_dir, max_rows=max_rows, tables=tables
),
}
for name in active:
if name in ("eda-hub", "bundle"):
continue
runner = runners.get(name)
if not runner:
continue
result = runner()
if name == "quick-stats":
qs = output_dir / "eda_quick_stats.md"
if qs.exists():
files.append(str(qs))
elif result is not None:
files.append(str(result))
hub = self.generate_eda_hub(output_dir)
files.append(str(hub))
return files, hub
def _bundle_quick_stats(
self,
db_path: Path,
output_dir: Path,
tables: Optional[List[str]],
) -> Optional[str]:
content = self.generate_quick_stats(db_path, tables=tables)
if content is None:
return None
path = output_dir / "eda_quick_stats.md"
path.write_text(content, encoding="utf-8")
return content