Historical Modelling

Navigation: ← Previous: DLS Resources | Next: Live Charts →

We train logistic regression and LightGBM on historical chase balls and compare against DLS.

import sys
from pathlib import Path
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
plt.style.use('seaborn-v0_8-whitegrid')

PROJ_DIR = Path('.').resolve()
if not (PROJ_DIR / 'cricsheet_utils.py').exists():
    PROJ_DIR = Path('projects/cricket-win-probability').resolve()
if str(PROJ_DIR) not in sys.path:
    sys.path.insert(0, str(PROJ_DIR))

from cricsheet_utils import (
    project_data_dir, load_cricsheet_json, match_metadata, balls_to_dataframe,
    build_game_states, cumulative_runs_by_innings, resource_remaining,
    dls_win_probability,
)
DATA_DIR = project_data_dir()
SAMPLE_DIR = DATA_DIR / 'sample_matches'

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import brier_score_loss, log_loss, roc_auc_score
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
import lightgbm as lgb

training = pd.read_csv(DATA_DIR / 'odi_training_sample.csv')
feature_cols = [
    'runs_scored', 'wickets_lost', 'balls_remaining', 'wickets_remaining',
    'runs_required', 'runs_required_per_ball', 'current_run_rate',
    'overs_completed', 'phase_powerplay', 'phase_death',
]
train_matches = training['match_idx'].unique()
rng = np.random.default_rng(42)
rng.shuffle(train_matches)
split = int(0.8 * len(train_matches))
train_ids, val_ids = set(train_matches[:split]), set(train_matches[split:])
train_df = training[training['match_idx'].isin(train_ids)]
val_df = training[training['match_idx'].isin(val_ids)]
X_train, y_train = train_df[feature_cols], train_df['batting_team_won']
X_val, y_val = val_df[feature_cols], val_df['batting_team_won']
print(f'Train balls: {len(X_train):,} | Val balls: {len(X_val):,}')

Train balls: 96,000 | Val balls: 24,000

Logistic regression

logit = Pipeline([
    ('scale', StandardScaler()),
    ('clf', LogisticRegression(max_iter=1000, random_state=42)),
])
logit.fit(X_train, y_train)
p_logit = logit.predict_proba(X_val)[:, 1]

LightGBM

lgbm = lgb.LGBMClassifier(
    n_estimators=200, learning_rate=0.05, max_depth=5,
    random_state=42, verbose=-1,
)
lgbm.fit(X_train, y_train)
p_lgbm = lgbm.predict_proba(X_val)[:, 1]

Metrics comparison

p_dls = val_df['dls_prob'].values
results = []
for name, preds in [('Logistic regression', p_logit), ('LightGBM', p_lgbm), ('DLS baseline', p_dls)]:
    results.append({
        'Model': name,
        'Brier score': round(brier_score_loss(y_val, preds), 4),
        'Log loss': round(log_loss(y_val, preds), 4),
        'AUC': round(roc_auc_score(y_val, preds), 4),
    })
metrics = pd.DataFrame(results)
metrics

	Model	Brier score	Log loss	AUC
0	Logistic regression	0.2510	0.6951	0.5065
1	LightGBM	0.2573	0.7117	0.5244
2	DLS baseline	0.3584	3.7750	0.4700

Logistic coefficients

coefs = pd.Series(logit.named_steps['clf'].coef_[0], index=feature_cols)
fig, ax = plt.subplots(figsize=(8, 4))
coefs.sort_values().plot(kind='barh', ax=ax, color='#1565c0')
ax.set_title('Logistic regression coefficients (scaled features)')
ax.set_xlabel('Coefficient')
plt.tight_layout()
plt.show()

import joblib
model_dir = PROJ_DIR / 'data'
joblib.dump(logit, model_dir / 'logit_model.joblib')
joblib.dump(lgbm, model_dir / 'lgbm_model.joblib')
print('Models saved.')

Models saved.

---

Navigation: ← Previous: DLS Resources | Next: Live Charts →