Spaces:
Runtime error
Runtime error
File size: 11,963 Bytes
993fce6 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 | """Clause Risk Classifier β Testing Script
========================================
Step 1: Run this ONCE to save the trained model to disk
Step 2: Use the saved model to:
(A) Predict on a new CSV file
(B) Type a single clause and get a prediction instantly
(C) Error analysis β see exactly where the model is wrong
"""
import os
import sys
import pickle
import argparse
import warnings
warnings.filterwarnings("ignore", category=UserWarning)
import numpy as np
import pandas as pd
import scipy.sparse as sp
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn.metrics import classification_report, confusion_matrix, recall_score
# ============================================================
# STEP 1 β SAVE YOUR TRAINED MODEL
# Run: python test_model.py --mode save --data "path/to/master_dataset.csv"
# This trains the model and saves it to model_bundle.pkl
# ============================================================
def save_model(data_path):
"""Train on full data and save everything needed for prediction."""
from sklearn.ensemble import VotingClassifier, ExtraTreesClassifier
import xgboost as xgb
import lightgbm as lgb
from sklearn.model_selection import train_test_split
print("Loading data...")
df = pd.read_csv(data_path)
num_cols = [
"modal_score", "consequence_score", "conditional_score",
"has_negation", "obligation_count", "penalty_flag", "word_count",
]
le = LabelEncoder()
y = le.fit_transform(df["risk_label"])
# Fit transformers on FULL dataset (for production use)
tfidf_word = TfidfVectorizer(max_features=2000, stop_words="english",
ngram_range=(1, 2), sublinear_tf=True)
tfidf_char = TfidfVectorizer(max_features=1000, analyzer="char_wb",
ngram_range=(3, 5), sublinear_tf=True)
scaler = StandardScaler()
text_word = tfidf_word.fit_transform(df["clean_text"].fillna(""))
text_char = tfidf_char.fit_transform(df["clean_text"].fillna(""))
num = sp.csr_matrix(scaler.fit_transform(df[num_cols].fillna(0).values))
X = sp.hstack([text_word, text_char, num], format="csr")
# Best params from v3 tuning
xgb_model = xgb.XGBClassifier(
objective="multi:softmax", num_class=len(le.classes_),
n_estimators=300, learning_rate=0.1, max_depth=8,
subsample=0.9, colsample_bytree=0.8, min_child_weight=1,
reg_alpha=0.0, reg_lambda=0.5, tree_method="hist",
random_state=42, eval_metric="mlogloss", verbosity=0,
)
lgb_model = lgb.LGBMClassifier(
objective="multiclass", num_class=len(le.classes_),
n_estimators=300, learning_rate=0.1, max_depth=8,
subsample=0.9, colsample_bytree=0.8, class_weight="balanced",
random_state=42, verbose=-1,
)
et_model = ExtraTreesClassifier(
n_estimators=300, min_samples_leaf=2,
class_weight="balanced", random_state=42, n_jobs=-1,
)
model = VotingClassifier(
estimators=[("xgb", xgb_model), ("lgb", lgb_model), ("et", et_model)],
voting="soft", n_jobs=1,
)
print("Training model on full dataset...")
model.fit(X, y)
# Bundle everything needed for inference
bundle = {
"model": model,
"tfidf_word": tfidf_word,
"tfidf_char": tfidf_char,
"scaler": scaler,
"le": le,
"num_cols": num_cols,
}
with open("model_bundle.pkl", "wb") as f:
pickle.dump(bundle, f)
print("\nModel saved to model_bundle.pkl")
print(f"Classes: {list(le.classes_)}")
# ============================================================
# SHARED HELPER β load bundle + transform raw input
# ============================================================
def load_bundle(path="model_bundle.pkl"):
if not os.path.exists(path):
print("ERROR: model_bundle.pkl not found.")
print("Run this first: python test_model.py --mode save --data your_data.csv")
sys.exit(1)
with open(path, "rb") as f:
return pickle.load(f)
def transform(bundle, texts, num_array):
"""Apply saved transformers to raw inputs."""
word = bundle["tfidf_word"].transform(texts)
char = bundle["tfidf_char"].transform(texts)
num = sp.csr_matrix(bundle["scaler"].transform(num_array))
return sp.hstack([word, char, num], format="csr")
# ============================================================
# MODE A β Predict on a new CSV file
# Run: python test_model.py --mode csv --data "path/to/new_file.csv"
# The CSV must have the same columns as the training data.
# If it has a risk_label column, a full evaluation report is shown.
# If not, predictions are just saved to predictions.csv.
# ============================================================
def test_on_csv(data_path):
bundle = load_bundle()
le = bundle["le"]
nc = bundle["num_cols"]
print(f"Loading: {data_path}")
df = pd.read_csv(data_path)
df["clean_text"] = df["clean_text"].fillna("")
X = transform(bundle, df["clean_text"], df[nc].fillna(0).values)
df["predicted_risk_label"] = le.inverse_transform(bundle["model"].predict(X))
# Confidence scores (probability of the winning class)
probs = bundle["model"].predict_proba(X)
df["confidence"] = (np.max(probs, axis=1) * 100).round(1).astype(str) + "%"
print("\n=== Sample Predictions (first 10) ===")
cols = ["clause_id", "clean_text", "predicted_risk_label", "confidence"]
if "risk_label" in df.columns:
cols.insert(3, "risk_label")
print(df.head(10)[cols].to_string(index=False))
# If ground-truth labels exist β full evaluation
if "risk_label" in df.columns:
y_true = le.transform(df["risk_label"])
y_pred = le.transform(df["predicted_risk_label"])
critical_idx = list(le.classes_).index("Critical")
print("\n=== Classification Report ===")
print(classification_report(df["risk_label"], df["predicted_risk_label"], digits=4))
per_class = recall_score(y_true, y_pred, average=None)
print(f"Critical Clause Recall: {per_class[critical_idx]:.4f}")
print("\nConfusion Matrix:")
cm = confusion_matrix(y_true, y_pred)
print(pd.DataFrame(cm, index=le.classes_, columns=le.classes_))
out_path = "predictions.csv"
df.to_csv(out_path, index=False)
print(f"\nSaved all predictions to {out_path}")
# ============================================================
# MODE B β Interactive: type a clause, get a prediction
# Run: python test_model.py --mode interactive
# Numeric features default to 0 β you can edit them in the prompt.
# ============================================================
def interactive_mode():
bundle = load_bundle()
le = bundle["le"]
nc = bundle["num_cols"]
severity_emoji = {"Critical": "CRITICAL", "High": "HIGH",
"Medium": "MEDIUM", "Low": "LOW"}
print("\n=== Interactive Clause Tester ===")
print("Type a clause and press Enter. Type 'quit' to exit.\n")
while True:
clause = input("Clause text: ").strip()
if clause.lower() in ("quit", "exit", "q"):
break
if not clause:
continue
# Optional: ask for numeric features
print("Numeric features (press Enter to use 0 for all):")
num_vals = []
for col in nc:
val = input(f" {col} [0]: ").strip()
try:
num_vals.append(float(val) if val else 0.0)
except ValueError:
num_vals.append(0.0)
X = transform(bundle, pd.Series([clause]),
np.array([num_vals]))
pred = bundle["model"].predict(X)[0]
probs = bundle["model"].predict_proba(X)[0]
label = le.inverse_transform([pred])[0]
print(f"\n Prediction : [{severity_emoji[label]}] {label}")
print(" Confidence breakdown:")
for i, cls in enumerate(le.classes_):
bar = "β" * int(probs[i] * 20)
print(f" {cls:<10s} {probs[i]*100:5.1f}% {bar}")
print()
# ============================================================
# MODE C β Error analysis
# Run: python test_model.py --mode errors --data "path/to/data.csv"
# Shows every clause the model got wrong, sorted by confidence.
# Helps you understand failure patterns.
# ============================================================
def error_analysis(data_path):
bundle = load_bundle()
le = bundle["le"]
nc = bundle["num_cols"]
df = pd.read_csv(data_path)
if "risk_label" not in df.columns:
print("ERROR: CSV must have a 'risk_label' column for error analysis.")
sys.exit(1)
df["clean_text"] = df["clean_text"].fillna("")
X = transform(bundle, df["clean_text"], df[nc].fillna(0).values)
y_true = le.transform(df["risk_label"])
y_pred = bundle["model"].predict(X)
probs = bundle["model"].predict_proba(X)
df["predicted_risk_label"] = le.inverse_transform(y_pred)
df["confidence"] = np.max(probs, axis=1).round(3)
df["correct"] = df["risk_label"] == df["predicted_risk_label"]
errors = df[~df["correct"]].copy()
errors = errors.sort_values("confidence", ascending=False)
total = len(df)
n_errors = len(errors)
print(f"\nTotal clauses : {total}")
print(f"Correct : {total - n_errors} ({(total-n_errors)/total*100:.1f}%)")
print(f"Wrong : {n_errors} ({n_errors/total*100:.1f}%)")
print("\n=== Most Confident Mistakes (model was very sure but wrong) ===")
top_errors = errors.head(20)[
["clause_id", "clean_text", "risk_label",
"predicted_risk_label", "confidence"]
]
pd.set_option("display.max_colwidth", 60)
print(top_errors.to_string(index=False))
print("\n=== Confusion breakdown (what gets confused with what) ===")
for true_cls in le.classes_:
subset = errors[errors["risk_label"] == true_cls]
if len(subset) == 0:
continue
counts = subset["predicted_risk_label"].value_counts()
print(f" True={true_cls:<10s} predicted as: "
+ ", ".join(f"{k}({v})" for k, v in counts.items()))
errors.to_csv("errors.csv", index=False)
print("\nAll errors saved to errors.csv")
# ============================================================
# MAIN
# ============================================================
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Test the clause risk classifier")
parser.add_argument("--mode", required=True,
choices=["save", "csv", "interactive", "errors"],
help="save | csv | interactive | errors")
parser.add_argument("--data",
default=r"C:\Users\Satya\Downloads\master_dataset.csv",
help="Path to CSV (needed for save / csv / errors modes)")
args = parser.parse_args()
if args.mode == "save":
data_path = args.data.strip().strip("\"'")
if not os.path.exists(data_path):
print(f"ERROR: File not found: {data_path}")
sys.exit(1)
save_model(data_path)
elif args.mode == "csv":
data_path = args.data.strip().strip("\"'")
if not os.path.exists(data_path):
print(f"ERROR: File not found: {data_path}")
sys.exit(1)
test_on_csv(data_path)
elif args.mode == "interactive":
interactive_mode()
elif args.mode == "errors":
data_path = args.data.strip().strip("\"'")
if not os.path.exists(data_path):
print(f"ERROR: File not found: {data_path}")
sys.exit(1)
error_analysis(data_path)
|