batch_91

Reviewed By: maggiemoss

Differential Revision: D101008119

Author: Pyre Bot Jr.

privacy_guard/analysis/code_similarity/code_bleu_node.py

@@ -30,8 +30,6 @@
 from privacy_guard.analysis.code_similarity.code_similarity_analysis_input import (
     CodeBleuAnalysisInput,
 )
-
-# pyre-ignore[21]: tree-sitter doesn't have properly exposed type stubs
 from tree_sitter import Node
 
 logger: logging.Logger = logging.getLogger(__name__)
@@ -73,7 +71,6 @@ def __init__(self, analysis_input: CodeBleuAnalysisInput) -> None:
         super().__init__(analysis_input=analysis_input)
 
     @staticmethod
-    # pyre-ignore[11]: Annotation `Node` is not defined as a type.
     def syntax_match(target_tree: Node, generated_tree: Node) -> float:
         def _node_sexp(node: Node) -> str:
             """Build a position-independent s-expression string for a subtree."""

privacy_guard/analysis/extraction/text_inclusion_analysis_node.py

@@ -106,7 +106,7 @@ def word_level_lcs_result_formatted(self) -> pd.DataFrame:
                     num_matched_words=num_matched_words,
                     matched_string=matched_string,
                     augmented_row=augmented_row,
-                    analysis_input=self.analysis_input,  # pyre-ignore
+                    analysis_input=self.analysis_input,
                 )
             )
 
@@ -166,7 +166,7 @@ def lcs_result_formatted(self, display_lcs_match: bool) -> pd.DataFrame:
                     lcs_dict=lcs_dict,
                     augmented_row=augmented_row,
                     display_lcs_match=display_lcs_match,
-                    analysis_input=self.analysis_input,  # pyre-ignore
+                    analysis_input=self.analysis_input,
                 )
             )
 

privacy_guard/analysis/lia/lia_analysis_input.py

@@ -22,11 +22,17 @@
 class LIAAnalysisInput(BaseAnalysisInput):
     def __init__(
         self,
+        # pyrefly: ignore [bad-specialization]
         predictions: NDArray[float],
+        # pyrefly: ignore [bad-specialization]
         predictions_y1_generation: NDArray[float],
+        # pyrefly: ignore [bad-specialization]
         true_bits: NDArray[int],
+        # pyrefly: ignore [bad-specialization]
         y0: NDArray[int],
+        # pyrefly: ignore [bad-specialization]
         y1: NDArray[int],
+        # pyrefly: ignore [bad-specialization]
         received_labels: NDArray[int],
     ) -> None:
         """

privacy_guard/analysis/lia/lia_analysis_node.py

@@ -125,8 +125,11 @@ def compute_scores(self, i: int) -> Tuple[torch.Tensor, torch.Tensor]:
             Tuple[torch.Tensor, torch.Tensor]: scores for samples with training labels and reconstructed labels
         """
 
+        # pyrefly: ignore [missing-attribute]
         received_labels = self._analysis_input.received_labels[i]
+        # pyrefly: ignore [missing-attribute]
         y1_probs = self._analysis_input.predictions_y1_generation
+        # pyrefly: ignore [missing-attribute]
         predictions = self._analysis_input.predictions
 
         if self.score_computation_function is not None:
@@ -142,6 +145,7 @@ def compute_scores(self, i: int) -> Tuple[torch.Tensor, torch.Tensor]:
                 np.log(prob_train + 1e-8) - np.log(prob_reconstruct + 1e-8)
             ) * prob_diff_label**self._power
 
+        # pyrefly: ignore [missing-attribute]
         true_bits = self._analysis_input.true_bits[i]
         scores_train = torch.tensor(scores[true_bits == 0])
         scores_test = torch.tensor(scores[true_bits == 1])
@@ -152,7 +156,9 @@ def run_analysis(self) -> BaseAnalysisOutput:
         """Run LIA analysis"""
 
         error_thresholds = np.linspace(0.01, 1, 100)
+        # pyrefly: ignore [missing-attribute]
         num_resampling = self._analysis_input.y1.shape[0]
+        # pyrefly: ignore [missing-attribute]
         num_samples = self._analysis_input.y1.shape[1]
 
         # run analysis for each game instance
@@ -221,9 +227,12 @@ def run_analysis(self) -> BaseAnalysisOutput:
             eps_at_tpr_bounds=(list(eps_tpr_lb), list(eps_tpr_ub)),
             eps_at_fpr_bounds=(list(eps_fpr_lb), list(eps_fpr_ub)),
             data_size=num_samples,
+            # pyrefly: ignore [missing-attribute]
             label_mean=np.mean(self._analysis_input.y0),
+            # pyrefly: ignore [missing-attribute]
             prediction_mean=np.mean(self._analysis_input.predictions),
             prediction_y1_generation_mean=np.mean(
+                # pyrefly: ignore [missing-attribute]
                 self._analysis_input.predictions_y1_generation
             ),
         )

privacy_guard/analysis/mia/analysis_node.py

@@ -235,6 +235,7 @@ def _calculate_one_off_eps(self) -> float:
         return eps_cp
 
     @staticmethod
+    # pyrefly: ignore [bad-specialization]
     def _compute_ci(array: NDArray[float], axis: int = 0) -> tuple[NDArray, NDArray]:
         """Compute confidence intervals (used for eps, auc, accuracy)"""
         # Sort along the specified axis
@@ -269,6 +270,7 @@ def _compute_bootstrap_sample_indexes(
         Returns:
             A list of indexes (with duplicates)
         """
+        # pyrefly: ignore [bad-return]
         return np.random.randint(0, num_users, sample_size)
 
     def run_analysis(self) -> BaseAnalysisOutput:

privacy_guard/analysis/mia/fdp_analysis_node.py

@@ -17,6 +17,8 @@
 import numpy as np
 from privacy_guard.analysis.base_analysis_node import BaseAnalysisNode
 from privacy_guard.analysis.base_analysis_output import BaseAnalysisOutput
+
+# pyrefly: ignore [missing-module-attribute]
 from scipy.stats import norm
 
 

privacy_guard/analysis/mia/fpr_lower_bound_analysis_node.py

@@ -65,14 +65,14 @@ class FPRLowerBoundAnalysisNodeOutput(BaseAnalysisOutput):
 
 
 def compute_metric_mean_with_ci(
+    # pyrefly: ignore [bad-specialization]
     metric_array: NDArray[float],
 ) -> tuple[float, float, float]:
     # TODO: Identify descriptive values for mean, lb, ub when bootstrap fails
 
     metric_mean = metric_array.mean()
     metric_mean_lb, metric_mean_ub = 0, 0
     try:
-        # pyre-ignore: Module `scipy.stats` has no attribute `bootstrap`.
         metric_mean_lb, metric_mean_ub = bootstrap(
             (metric_array,), statistic=np.mean, method="BCa"
         ).confidence_interval
@@ -181,6 +181,7 @@ def run_analysis(self) -> BaseAnalysisOutput:
 
         return outputs
 
+    # pyrefly: ignore [bad-specialization]
     def _make_acc_auc_epsilon_array(self) -> NDArray[float]:
         """
         Make list of tuples metrics at error thresholds, each of which contains the

privacy_guard/analysis/mia/mia_results.py

@@ -15,6 +15,8 @@
 import numpy as np
 import torch
 from numpy.typing import NDArray
+
+# pyrefly: ignore [missing-module-attribute]
 from scipy.stats import beta
 from sklearn.metrics import auc, roc_curve
 
@@ -41,9 +43,12 @@ def __init__(self, scores_train: torch.Tensor, scores_test: torch.Tensor) -> Non
 
     def _get_indices_of_error_at_thresholds(
         self,
+        # pyrefly: ignore [bad-specialization]
         error_rates: NDArray[float],
+        # pyrefly: ignore [bad-specialization]
         error_thresholds: NDArray[float],
         error_type: str,
+        # pyrefly: ignore [bad-specialization]
     ) -> NDArray[int]:
         """
         Get indices where error values are greater/smaller than error thresholds.
@@ -80,6 +85,7 @@ def _get_indices_of_error_at_thresholds(
         else:
             raise ValueError(f"Invalid error type: {error_type}")
 
+    # pyrefly: ignore [bad-specialization]
     def get_tpr_fpr(self) -> tuple[NDArray[float], NDArray[float]]:
         """
         Computes true positive rate and true negative rate given scores and labels indicating membership.
@@ -213,6 +219,7 @@ def compute_acc_auc_ci_epsilon(self, delta: float) -> tuple[float, float, float]
     def compute_metrics_at_error_threshold(
         self,
         delta: float,
+        # pyrefly: ignore [bad-specialization]
         error_threshold: NDArray[float],
         cap_eps: bool = True,
         verbose: bool = False,

privacy_guard/analysis/mia/parallel_analysis_node.py

@@ -123,6 +123,7 @@ def _compute_metrics_array(
                 f"An exception occurred when computing acc/auc/epsilon metrics: {e}"
             )
 
+        # pyrefly: ignore [bad-return]
         return metrics_results
 
     def _parallel_compute_chunk_sizes(self, task_num: int) -> list[int]:

privacy_guard/analysis/mia/parallel_fpr_lower_bound_analysis_node.py

@@ -40,7 +40,6 @@ def compute_metric_mean_with_ci(
     metric_mean = metric_array.mean()
     metric_mean_lb, metric_mean_ub = 0, 0
     try:
-        # pyre-ignore: Module `scipy.stats` has no attribute `bootstrap`.
         metric_mean_lb, metric_mean_ub = bootstrap(
             (metric_array,), statistic=np.mean, method="BCa"
         ).confidence_interval
@@ -128,6 +127,7 @@ def _compute_metrics_and_eps_fpr_array(
                 f"An exception occurred when computing acc/auc/epsilon metrics: {e}"
             )
 
+        # pyrefly: ignore [bad-return]
         return metrics_results, eps_fpr_results
 
     def _parallel_compute_chunk_sizes(self, task_num: int) -> list[int]: