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