Whitespace & Pre-Commit

.pre-commit-config.yaml

@@ -0,0 +1,6 @@
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.6.0  # Use the latest stable version
+    hooks:
+      - id: trailing-whitespace
+

README.md

@@ -4,13 +4,13 @@
 [![PyPI Downloads](https://img.shields.io/pypi/dm/tippingpt.svg?label=PyPI%20downloads)](
 https://pypi.org/project/tippingpt/)
 
-A lightweight, high-performance marketing intelligence module that uses machine learning and calculus to determine the exact inflection points of a media response curve. 
+A lightweight, high-performance marketing intelligence module that uses machine learning and calculus to determine the exact inflection points of a media response curve.
 
 Growth marketers and media buyers constantly ask two questions: *"When are we out of the inefficient learning phase?"* and *"When should we stop scaling spend?"* By fitting historical performance data to a continuous mathematical curve, this tool identifies the **Minimal Marginal Cost Point** (where efficiency peaks) and the **Point of Diminishing Returns** (where scaling is no longer profitable), defining your exact **Optimal Scaling Zone**.
 
 ## 🧠 Methodology
 
-This project leverages the mathematical foundations of modern Marketing Mix Modeling (MMM)—specifically the techniques popularized by [Google’s Meridian](https://github.com/google/meridian). 
+This project leverages the mathematical foundations of modern Marketing Mix Modeling (MMM)—specifically the techniques popularized by [Google’s Meridian](https://github.com/google/meridian).
 
 Instead of basic linear or logarithmic approximations, this module natively models media saturation using the **Hill Function**.
 

pyproject.toml

@@ -7,7 +7,7 @@ keywords = [
     "tipping point",
     "marketing",
     "utility",
-    "cli" 
+    "cli"
 ]
 
 classifiers = [
@@ -16,7 +16,7 @@ classifiers = [
     "Programming Language :: Python :: 3.12",
     "Programming Language :: Python :: 3.13",
     "Operating System :: OS Independent",
-    "Development Status :: 3 - Alpha", 
+    "Development Status :: 3 - Alpha",
 ]
 authors = [
     { name = "Ryan Duecker", email = "ryanduecker@gmail.com" },

src/tippingpoint/curve.py

@@ -65,22 +65,22 @@ def log_posterior(params):
         np.std(y) * 0.1
       ])
       current_log_post = log_posterior(current_params)
-      
+
       samples = []
       # Adaptive step size (simplified)
       step_size = current_params * 0.05
 
       for i in range(n_samples + burn_in):
         proposal = current_params + np.random.normal(0, step_size)
         proposal_log_post = log_posterior(proposal)
-        
+
         if proposal_log_post > current_log_post or np.random.rand() < np.exp(proposal_log_post - current_log_post):
           current_params = proposal
           current_log_post = proposal_log_post
-        
+
         if i >= burn_in:
           samples.append(current_params.copy())
-          
+
         # Small adaptation during burn-in
         if i < burn_in and i % 100 == 0 and i > 0:
           # This is a very crude adaptation
@@ -95,12 +95,12 @@ def log_posterior(params):
       'K': posterior[:, 2],
       'sigma': posterior[:, 3]
     }
-    
+
     # Point estimates (posterior mean)
     beta_mean = np.mean(samples_dict['beta'])
     alpha_mean = np.mean(samples_dict['alpha'])
     K_mean = np.mean(samples_dict['K'])
-    
+
     print(f"[{channel_name}] Bayesian fit complete. Samples: {len(posterior)}")
     return cls(beta_mean, alpha_mean, K_mean, channel_name, posterior_samples=samples_dict)
 
@@ -218,13 +218,13 @@ def plot_response_curve(self, target_mroas=1.0, current_spend=None, show_interva
     plot_limit = max(plot_limit, current_spend * 1.2 if current_spend else 0)
 
     x_vals = np.linspace(0, plot_limit, 500)
-    
+
     if show_intervals and self.posterior_samples:
       y_returns_dist = self.predict_incremental_return(x_vals, use_samples=True)
       y_return = np.mean(y_returns_dist, axis=0)
       y_return_low = np.percentile(y_returns_dist, 5, axis=0)
       y_return_high = np.percentile(y_returns_dist, 95, axis=0)
-      
+
       y_mroas_dist = self.predict_marginal_return(x_vals, use_samples=True)
       y_mroas = np.mean(y_mroas_dist, axis=0)
       y_mroas_low = np.percentile(y_mroas_dist, 5, axis=0)
@@ -238,18 +238,18 @@ def plot_response_curve(self, target_mroas=1.0, current_spend=None, show_interva
     ax1.plot(x_vals, y_return, color='#2CA02C', linewidth=3, label="Incremental Return")
     if show_intervals and self.posterior_samples:
       ax1.fill_between(x_vals, y_return_low, y_return_high, color='#2CA02C', alpha=0.2, label="90% Credible Interval")
-    
+
     ax1.set_xlabel('Spend ($)', fontsize=12, fontweight='bold')
     ax1.set_ylabel('Incremental Return', color='#2CA02C', fontsize=12, fontweight='bold')
     ax1.tick_params(axis='y', labelcolor='#2CA02C')
     ax1.grid(True, linestyle='--', alpha=0.5)
-    
+
     # Secondary Axis: Marginal Return
     ax2 = ax1.twinx()
     ax2.plot(x_vals, y_mroas, color='#1F77B4', linestyle='--', linewidth=2, label="Marginal ROAS")
     if show_intervals and self.posterior_samples:
       ax2.fill_between(x_vals, y_mroas_low, y_mroas_high, color='#1F77B4', alpha=0.1)
-    
+
     ax2.set_ylabel('Marginal ROAS (mROAS)', color='#1F77B4', fontsize=12, fontweight='bold')
     ax2.tick_params(axis='y', labelcolor='#1F77B4')
     ax2.axhline(target_mroas, color='gray', linestyle=':', label="Target mROAS Floor")

tests/test_bayesian.py

@@ -17,7 +17,7 @@ def synthetic_data():
 def test_fit_bayesian_basic(synthetic_data):
     x, y = synthetic_data
     model = MarketingReturnCurve.fit_bayesian(x, y, n_samples=500, burn_in=100, chains=2)
-    
+
     assert model.beta > 0
     assert model.alpha > 0
     assert model.K > 0
@@ -33,7 +33,7 @@ def test_fit_bayesian_with_priors(synthetic_data):
         'K': (np.log(20000), 0.1)
     }
     model = MarketingReturnCurve.fit_bayesian(x, y, priors=priors, n_samples=200, burn_in=50, chains=1)
-    
+
     # Check if results are close to true values due to tight priors
     assert 90000 < model.beta < 110000
     assert 1.3 < model.alpha < 1.7
@@ -42,11 +42,11 @@ def test_fit_bayesian_with_priors(synthetic_data):
 def test_predict_with_samples(synthetic_data):
     x, y = synthetic_data
     model = MarketingReturnCurve.fit_bayesian(x, y, n_samples=100, burn_in=50, chains=1)
-    
+
     # Incremental return
     preds = model.predict_incremental_return([1000, 2000], use_samples=True)
     assert preds.shape == (100, 2)
-    
+
     # Marginal return
     m_preds = model.predict_marginal_return([1000, 2000], use_samples=True)
     assert m_preds.shape == (100, 2)

tests/test_curve.py

@@ -12,7 +12,7 @@ def setup_method(self):
     """Setup a baseline S-Curve for use in multiple tests."""
     # Parameters: Max Return=10000, Shape=2.0 (S-Curve), Half-Saturation=500
     self.s_curve = MarketingReturnCurve(beta=10000.0, alpha=2.0, half_saturation_k=500.0, channel_name="Test_S_Curve")
-    
+
     # Parameters: Shape=0.8 (C-Curve, no warm-up phase)
     self.c_curve = MarketingReturnCurve(beta=10000.0, alpha=0.8, half_saturation_k=500.0, channel_name="Test_C_Curve")
 
@@ -45,7 +45,7 @@ def test_minimal_marginal_cost_point_s_curve(self):
     """Test f''(x) = 0 for an S-Curve (alpha > 1)."""
     expected_inflection = 500.0 * np.sqrt(1.0 / 3.0) # For alpha = 2.0, inflection = K * sqrt((2-1)/(2+1)) = 500 * sqrt(1/3) ≈ 288.675
     actual_inflection = self.s_curve.get_minimal_marginal_cost_point()
-    
+
     assert actual_inflection == pytest.approx(expected_inflection, rel=1e-3)
 
   def test_minimal_marginal_cost_point_c_curve(self):
@@ -56,7 +56,7 @@ def test_diminishing_returns_point_valid(self):
     """Test finding the target spend level for a reachable mROAS."""
     target_mroas = 5.0
     spend_cap = self.s_curve.get_diminishing_returns_point(target_mroas)
-    
+
     assert spend_cap is not None
     actual_mroas_at_cap = self.s_curve.predict_marginal_return(spend_cap)
     assert actual_mroas_at_cap == pytest.approx(target_mroas, rel=1e-3)
@@ -75,7 +75,7 @@ def test_tinygrad_optimization_engine(self):
     returns = (50000 * spends**1.5) / (4000**1.5 + spends**1.5) # Fake responses following roughly an S curve
     # Run with very few epochs just to verify the math/graph builds and executes properly
     model = MarketingReturnCurve.from_historical_data( spend_array=spends, return_array=returns, epochs=10, lr=0.1)# Fast execution for test suite
-    
+
     assert isinstance(model, MarketingReturnCurve)
     assert model.beta > 0
     assert model.alpha > 0