style: remove Japanese docstrings and inline comments, use English only

Author: YuminosukeSato

python/causal_impact/analysis.py

@@ -11,45 +11,34 @@
 class CausalImpactResults:
     """Results of causal impact analysis."""
 
-    # Observed data
-    actual: np.ndarray  # (T_post,) observed y in post period
-
-    # Pointwise effects
-    point_effects: np.ndarray  # (T_post,) mean effect per time point
-    point_effect_lower: np.ndarray  # (T_post,) lower CI per time point
-    point_effect_upper: np.ndarray  # (T_post,) upper CI per time point
-    ci_lower: float  # lower CI bound on average effect
-    ci_upper: float  # upper CI bound on average effect
-    point_effect_mean: float  # mean of point effects across time
-    average_effect_sd: float  # std of per-sample average effects
-
-    # Cumulative effects
-    cumulative_effect: np.ndarray  # (T_post,) cumulative point effects
-    cumulative_effect_lower: np.ndarray  # (T_post,) lower cumulative CI
-    cumulative_effect_upper: np.ndarray  # (T_post,) upper cumulative CI
-    cumulative_effect_total: float  # total cumulative effect
-    cumulative_effect_sd: float  # std of per-sample cumulative effects
-
-    # Relative effects
-    relative_effect_mean: float  # relative effect (effect / predicted)
-    relative_effect_sd: float  # std of per-sample relative effects
-    relative_effect_lower: float  # lower CI on relative effect
-    relative_effect_upper: float  # upper CI on relative effect
-
-    # Significance
-    p_value: float  # Bayesian one-sided tail probability
-
-    # Counterfactual predictions
-    predictions_mean: np.ndarray  # (T_post,) mean counterfactual
-    predictions_sd: np.ndarray  # (T_post,) std of predictions per time point
-    predictions_lower: np.ndarray  # (T_post,) lower CI counterfactual
-    predictions_upper: np.ndarray  # (T_post,) upper CI counterfactual
-    average_prediction_sd: float  # std of per-sample average predictions
-    average_prediction_lower: float  # lower CI on average prediction
-    average_prediction_upper: float  # upper CI on average prediction
-    cumulative_prediction_sd: float  # std of per-sample cumulative predictions
-    cumulative_prediction_lower: float  # lower CI on cumulative prediction
-    cumulative_prediction_upper: float  # upper CI on cumulative prediction
+    actual: np.ndarray
+    point_effects: np.ndarray
+    point_effect_lower: np.ndarray
+    point_effect_upper: np.ndarray
+    ci_lower: float
+    ci_upper: float
+    point_effect_mean: float
+    average_effect_sd: float
+    cumulative_effect: np.ndarray
+    cumulative_effect_lower: np.ndarray
+    cumulative_effect_upper: np.ndarray
+    cumulative_effect_total: float
+    cumulative_effect_sd: float
+    relative_effect_mean: float
+    relative_effect_sd: float
+    relative_effect_lower: float
+    relative_effect_upper: float
+    p_value: float
+    predictions_mean: np.ndarray
+    predictions_sd: np.ndarray
+    predictions_lower: np.ndarray
+    predictions_upper: np.ndarray
+    average_prediction_sd: float
+    average_prediction_lower: float
+    average_prediction_upper: float
+    cumulative_prediction_sd: float
+    cumulative_prediction_lower: float
+    cumulative_prediction_upper: float
 
 
 class CausalAnalysis:
@@ -66,28 +55,19 @@ def compute_effects(
 
         n_samples = predictions.shape[0]
 
-        # Effect per sample per time point: observed - counterfactual
-        # predictions shape: (n_samples, t_post)
-        effects = y_post[np.newaxis, :] - predictions  # (n_samples, t_post)
+        effects = y_post[np.newaxis, :] - predictions
+        avg_effects = effects.mean(axis=1)
+        point_effects = effects.mean(axis=0)
 
-        # Average effect across time for each sample
-        avg_effects = effects.mean(axis=1)  # (n_samples,)
-
-        # Point effects: mean across samples at each time point
-        point_effects = effects.mean(axis=0)  # (t_post,)
-
-        # Summary-table CI on average effect uses sample-average quantiles.
         lower_q = alpha / 2
         upper_q = 1 - alpha / 2
         point_effect_lower = np.percentile(effects, 100 * lower_q, axis=0)
         point_effect_upper = np.percentile(effects, 100 * upper_q, axis=0)
         ci_lower = float(np.percentile(avg_effects, 100 * lower_q))
         ci_upper = float(np.percentile(avg_effects, 100 * upper_q))
 
-        # Mean effect
         point_effect_mean = float(avg_effects.mean())
 
-        # Cumulative effect
         cumulative_effect = np.cumsum(point_effects)
         cum_effects_samples = np.cumsum(effects, axis=1)
         cumulative_effect_lower = np.percentile(
@@ -102,18 +82,15 @@ def compute_effects(
         )
         cumulative_effect_total = float(cumulative_effect[-1])
 
-        # Actual observed values
         actual = y_post.copy()
 
-        # Per-time-point std of predictions across samples
         if n_samples == 1:
             predictions_sd_arr = np.zeros(predictions.shape[1])
         else:
             predictions_sd_arr = np.std(predictions, axis=0, ddof=1)
 
-        # Prediction scalars (cross-sample aggregates)
-        avg_pred_per_sample = predictions.mean(axis=1)  # (n_samples,)
-        cum_pred_per_sample = predictions.sum(axis=1)  # (n_samples,)
+        avg_pred_per_sample = predictions.mean(axis=1)
+        cum_pred_per_sample = predictions.sum(axis=1)
 
         if n_samples == 1:
             average_prediction_sd = 0.0
@@ -135,8 +112,7 @@ def compute_effects(
             np.percentile(cum_pred_per_sample, 100 * upper_q)
         )
 
-        # Effect s.d. scalars
-        cum_effects_per_sample = effects.sum(axis=1)  # (n_samples,)
+        cum_effects_per_sample = effects.sum(axis=1)
 
         if n_samples == 1:
             average_effect_sd = 0.0
@@ -145,7 +121,6 @@ def compute_effects(
             average_effect_sd = float(np.std(avg_effects, ddof=1))
             cumulative_effect_sd = float(np.std(cum_effects_per_sample, ddof=1))
 
-        # Relative effect per sample
         avg_pred_per_sample_safe = np.where(
             np.abs(avg_pred_per_sample) > 1e-10,
             avg_pred_per_sample,
@@ -171,15 +146,12 @@ def compute_effects(
             np.percentile(rel_effects_per_sample, 100 * upper_q)
         )
 
-        # p-value: proportion of samples where average effect has opposite sign
         if point_effect_mean >= 0:
             p_value = float(np.mean(avg_effects < 0))
         else:
             p_value = float(np.mean(avg_effects > 0))
-        # Ensure minimum p-value of 1/n_samples
         p_value = max(p_value, 1.0 / n_samples)
 
-        # Counterfactual prediction summaries
         predictions_mean = predictions.mean(axis=0)
         predictions_lower = np.percentile(predictions, 100 * lower_q, axis=0)
         predictions_upper = np.percentile(predictions, 100 * upper_q, axis=0)

python/causal_impact/summary.py

@@ -6,23 +6,20 @@
 
 
 class SummaryFormatter:
-    """Format CausalImpact results as text summary or natural language report."""
+    """Format CausalImpact results as text summary or report."""
 
     @staticmethod
     def summary(results: CausalImpactResults, digits: int = 2) -> str:
         fmt = f".{digits}f"
 
-        # Actual
         avg_actual = format(results.actual.mean(), fmt)
         cum_actual = format(results.actual.sum(), fmt)
 
-        # Prediction
         avg_pred = format(results.predictions_mean.mean(), fmt)
         avg_pred_sd = format(results.average_prediction_sd, fmt)
         cum_pred = format(results.predictions_mean.sum(), fmt)
         cum_pred_sd = format(results.cumulative_prediction_sd, fmt)
 
-        # Prediction CI
         avg_pred_ci = (
             f"[{format(results.average_prediction_lower, fmt)}, "
             f"{format(results.average_prediction_upper, fmt)}]"
@@ -32,13 +29,11 @@ def summary(results: CausalImpactResults, digits: int = 2) -> str:
             f"{format(results.cumulative_prediction_upper, fmt)}]"
         )
 
-        # Absolute effect
         avg_eff = format(results.point_effect_mean, fmt)
         avg_eff_sd = format(results.average_effect_sd, fmt)
         cum_eff = format(results.cumulative_effect_total, fmt)
         cum_eff_sd = format(results.cumulative_effect_sd, fmt)
 
-        # Absolute effect CI
         avg_eff_ci = (
             f"[{format(results.ci_lower, fmt)}, {format(results.ci_upper, fmt)}]"
         )
@@ -47,7 +42,6 @@ def summary(results: CausalImpactResults, digits: int = 2) -> str:
             f"{format(results.cumulative_effect_upper[-1], fmt)}]"
         )
 
-        # Relative effect
         rel_m = format(results.relative_effect_mean * 100, fmt)
         rel_sd = format(results.relative_effect_sd * 100, fmt)
         rel_lo = format(results.relative_effect_lower * 100, fmt)