File size: 9,353 Bytes
a7284df |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 |
import io, math, json, gzip, textwrap
import numpy as np
import pandas as pd
import gradio as gr
from typing import Dict, Any
# --- (Functions below are minimal clones to keep the Gradio app standalone) ---
def shannon_entropy_from_counts(counts: np.ndarray) -> float:
counts = counts.astype(float)
total = counts.sum()
if total <= 0:
return 0.0
p = counts / total
p = p[p > 0]
return float(-(p * np.log2(p)).sum())
def numeric_binned_entropy(series: pd.Series, bins: int = 32):
x = series.dropna().astype(float).values
if x.size == 0:
return 0.0, 0
try:
qs = np.linspace(0, 1, bins + 1)
edges = np.unique(np.nanpercentile(x, qs * 100))
if len(edges) < 2:
edges = np.unique(x)
hist, _ = np.histogram(x, bins=edges)
except Exception:
hist, _ = np.histogram(x, bins=bins)
H = shannon_entropy_from_counts(hist)
k = np.count_nonzero(hist)
return H, max(k, 1)
def categorical_entropy(series: pd.Series):
x = series.dropna().astype(str).values
if x.size == 0:
return 0.0, 0
vals, counts = np.unique(x, return_counts=True)
H = shannon_entropy_from_counts(counts)
return H, len(vals)
def monotone_runs_and_entropy(series: pd.Series):
x = series.dropna().values
n = len(x)
if n <= 1:
return 1, 0.0
runs = [1]
for i in range(1, n):
if x[i] >= x[i-1]:
runs[-1] += 1
else:
runs.append(1)
run_lengths = np.array(runs, dtype=float)
H = shannon_entropy_from_counts(run_lengths)
return len(runs), H
def sortedness_score(series: pd.Series) -> float:
x = series.dropna().values
if len(x) <= 1:
return 1.0
return float(np.mean(np.diff(x) >= 0))
def gzip_compress_ratio_from_bytes(b: bytes) -> float:
if len(b) == 0:
return 1.0
out = io.BytesIO()
with gzip.GzipFile(fileobj=out, mode="wb") as f:
f.write(b)
compressed = out.getvalue()
return len(compressed) / len(b)
def dataframe_gzip_ratio(df: pd.DataFrame, max_rows: int = 20000) -> float:
s = df.sample(min(len(df), max_rows), random_state=0) if len(df) > max_rows else df
raw = s.to_csv(index=False).encode("utf-8", errors="ignore")
return gzip_compress_ratio_from_bytes(raw)
def pareto_maxima_count(points: np.ndarray) -> int:
if points.shape[1] < 2 or points.shape[0] == 0:
return 0
P = points[:, :2]
order = np.lexsort((-P[:, 1], -P[:, 0]))
best_y = -np.inf
count = 0
for idx in order:
y = P[idx, 1]
if y >= best_y:
count += 1
best_y = y
return int(count)
def kd_entropy(points: np.ndarray, max_leaf: int = 128, axis: int = 0) -> float:
n = points.shape[0]
if n == 0:
return 0.0
if n <= max_leaf:
return 0.0
d = points.shape[1]
vals = points[:, axis]
med = np.median(vals)
left = points[vals <= med]
right = points[vals > med]
pL = len(left) / n
pR = len(right) / n
H_here = 0.0
for p in (pL, pR):
if p > 0:
H_here += -p * math.log(p, 2)
next_axis = (axis + 1) % d
return H_here + kd_entropy(left, max_leaf, next_axis) + kd_entropy(right, max_leaf, next_axis)
def normalize(value: float, max_value: float) -> float:
if max_value <= 0:
return 0.0
v = max(0.0, min(1.0, value / max_value))
return float(v)
def compute_metrics(df: pd.DataFrame):
report = {}
n_rows, n_cols = df.shape
report["shape"] = {"rows": int(n_rows), "cols": int(n_cols)}
# Types
types = {}
for c in df.columns:
s = df[c]
if pd.api.types.is_numeric_dtype(s):
types[c] = "numeric"
elif pd.api.types.is_datetime64_any_dtype(s) or "date" in str(s.dtype).lower():
types[c] = "datetime"
else:
types[c] = "categorical"
report["column_types"] = types
missing = df.isna().mean().to_dict()
dup_ratio = float((len(df) - len(df.drop_duplicates())) / max(1, len(df)))
report["missing_fraction_per_column"] = {k: float(v) for k, v in missing.items()}
report["duplicate_row_fraction"] = dup_ratio
col_stats = {}
for c in df.columns:
s = df[c]
if types[c] == "numeric":
H, k = numeric_binned_entropy(s)
runs, Hruns = monotone_runs_and_entropy(s)
sorted_frac = sortedness_score(s)
col_stats[c] = {
"entropy_binned_bits": float(H),
"active_bins": int(k),
"monotone_runs": int(runs),
"run_entropy_bits": float(Hruns),
"sortedness_fraction": float(sorted_frac),
}
else:
H, k = categorical_entropy(s)
col_stats[c] = {"entropy_bits": float(H), "unique_values": int(k)}
report["per_column"] = col_stats
try:
gzip_ratio = dataframe_gzip_ratio(df)
except Exception:
gzip_ratio = 1.0
report["gzip_compression_ratio"] = float(gzip_ratio)
num_cols = [c for c, t in types.items() if t == "numeric"]
if len(num_cols) >= 2:
X = df[num_cols].select_dtypes(include=[np.number]).values.astype(float)
X = X[~np.isnan(X).any(axis=1)]
if X.shape[0] >= 3:
pts2 = X[:, :2]
report["pareto_maxima_2d"] = int(pareto_maxima_count(pts2))
try:
H_kd = kd_entropy(pts2, max_leaf=128, axis=0)
except Exception:
H_kd = 0.0
report["kd_partition_entropy_bits"] = float(H_kd)
else:
report["pareto_maxima_2d"] = 0
report["kd_partition_entropy_bits"] = 0.0
else:
report["pareto_maxima_2d"] = 0
report["kd_partition_entropy_bits"] = 0.0
max_bits = math.log2(max(2, n_rows))
he_parts = []
he_parts.append(1.0 - max(0.0, min(1.0, report["gzip_compression_ratio"])))
num_run_entropies = []
for c in df.columns:
st = col_stats.get(c, {})
if "run_entropy_bits" in st:
num_run_entropies.append(st["run_entropy_bits"])
if num_run_entropies:
mean_run_H = float(np.mean(num_run_entropies))
he_parts.append(1.0 - normalize(mean_run_H, max_bits))
H_kd = report.get("kd_partition_entropy_bits", 0.0)
if H_kd is not None:
he_parts.append(1.0 - normalize(float(H_kd), max_bits))
if he_parts:
HE = float(np.mean([max(0.0, min(1.0, v)) for v in he_parts]))
else:
HE = 0.0
report["harvestable_energy_score"] = HE
return report
def explain_report(report: Dict[str, Any]) -> str:
lines = []
r, c = report["shape"]["rows"], report["shape"]["cols"]
lines.append(f"**Dataset shape:** {r} rows × {c} columns.")
g = report.get("gzip_compression_ratio", None)
if g is not None:
lines.append(f"**Global compressibility (gzip ratio):** {g:.3f}. Lower = more structure.")
he = report.get("harvestable_energy_score", 0.0)
he_pct = int(100 * he)
lines.append(f"**Harvestable Energy (0–100):** ~{he_pct}. Higher = more exploitable order.")
pm = report.get("pareto_maxima_2d", None)
if pm is not None:
lines.append(f"**2D Pareto maxima (first two numeric cols):** {pm}.")
Hkd = report.get("kd_partition_entropy_bits", None)
if Hkd is not None:
lines.append(f"**Range-partition entropy (kd approx):** {Hkd:.3f} bits.")
lines.append("\\n**Column-level:**")
for c, st in report.get("per_column", {}).items():
m = report["missing_fraction_per_column"].get(c, 0.0)
if "entropy_binned_bits" in st:
lines.append(f"- **{c}** (numeric): missing {m:.1%}, binned entropy {st['entropy_binned_bits']:.2f} bits, "
f"{st['monotone_runs']} runs (run-entropy {st['run_entropy_bits']:.2f} bits), "
f"sortedness {st['sortedness_fraction']:.2f}.")
elif "entropy_bits" in st:
lines.append(f"- **{c}** (categorical): missing {m:.1%}, entropy {st['entropy_bits']:.2f} bits, "
f"{st['unique_values']} unique.")
else:
lines.append(f"- **{c}**: missing {m:.1%}.")
lines.append("\\n**Tips:** Higher energy and lower entropies often allow near-linear algorithms (run-aware sorts, hull scans, envelope merges).")
return "\\n".join(lines)
def analyze(file):
if file is None:
return "Please upload a CSV.", ""
try:
df = pd.read_csv(file.name)
except Exception as e:
return f"Failed to read CSV: {e}", ""
report = compute_metrics(df)
md = explain_report(report)
return json.dumps(report, indent=2), md
with gr.Blocks(title="Dataset Energy & Entropy Analyzer") as demo:
gr.Markdown("# Dataset Energy & Entropy Analyzer\nUpload a CSV to compute dataset structure metrics (entropy, runs, compressibility, kd-entropy) and an overall **Harvestable Energy** score.")
with gr.Row():
inp = gr.File(file_types=[".csv"], label="CSV file")
with gr.Row():
btn = gr.Button("Analyze", variant="primary")
with gr.Row():
json_out = gr.Code(label="Raw report (JSON)", language="json")
md_out = gr.Markdown()
btn.click(analyze, inputs=inp, outputs=[json_out, md_out])
if __name__ == "__main__":
demo.launch() |