Jean/efficient bootstrapping (#3)

* trying

* not working because of weird class vs instance issue

* getting nowhere

* errors start to make more sense

* implicit self

* supposedly working if no reinstantiation

* weird bugs still...

* fixing inii of algo

* iterating

* making slow progress

* it runs !

* fixing criterrion

* fixing predict

* fixing metric

* more robust equality check but is it needed ?

* more strength to the popensity to see actual learning

* starting up tests

* trying to launch tests

* first CP of tests run

* tests don't pass hmm...

* trying to fix bootstrap

* bootstrapping is working

* trying to make linting better

* trying to make linting better

* trying to make linting better

* trying to make linting better

* trying to make linting better

* adding better docstring

* trying to make docs pass

* trying to make docs pass

* adding doc and fixing it

* trying to make sphinx pass

* trying to fix CI

* tryiing to make CI pass

* upping version

* trying to install the correct version

* fixing data error

* first test working

* fixing linting

* accomodating new update_from_checkpoint functon

* Update setup.py

* accomodating substrafl api chhange

* new API compliance

docs/source/api/strategies.rst

@@ -5,4 +5,8 @@ fedeca.strategies
 
 .. autoclass:: fedeca.strategies.WebDisco
 
+.. automodule:: fedeca.strategies.bootstraper
+
 .. automodule:: fedeca.strategies.webdisco_utils
+
+

fedeca/__init__.py

@@ -1,3 +1,4 @@
 """Top level package for :mod:`fedeca`."""
 from .fedeca_core import FedECA
+from .fedeca_core import LogisticRegressionTorch
 from .competitors import PooledIPTW, MatchingAjudsted, NaiveComparison

fedeca/algorithms/torch_dp_fed_avg_algo.py

@@ -1,5 +1,6 @@
 """Differentially private algorithm to be used with FedAvg strategy."""
 import logging
+import random
 from typing import Any, Optional
 
 import numpy as np
@@ -371,28 +372,26 @@ def _get_state_to_save(self) -> dict:
 
         return checkpoint
 
-    def _update_from_checkpoint(self, path) -> dict:
+    def _update_from_checkpoint(self, checkpoint: dict) -> None:
         """Set self attributes using saved values.
 
         Parameters
         ----------
-        path : Path
-            Path towards the checkpoint to use.
+        checkpoint : dict
+            Checkpoint to load.
 
         Returns
         -------
         dict
             The emptied checkpoint.
         """
-        # One cannot simply call checkpoint = super()._update_from_checkpoint(path)
+        # One cannot simply call checkpoint = super()._update_from_checkpoint(chkpt)
         # because we have to change the model class if it should be changed
         # (and optimizer) aka if we find a specific key in the checkpoint
-        assert (
-            path.is_file()
-        ), f'Cannot load the model - does not exist {list(path.parent.glob("*"))}'
-        checkpoint = torch.load(path, map_location=self._device)
+
         # For some reason substrafl save and load client before calling train
         if "privacy_accountant_state_dict" in checkpoint:
+
             self.accountant = RDPAccountant()
             self.accountant.load_state_dict(
                 checkpoint.pop("privacy_accountant_state_dict")
@@ -429,10 +428,13 @@ def _update_from_checkpoint(self, path) -> dict:
 
         self._index_generator = checkpoint.pop("index_generator")
 
+        random.setstate(checkpoint.pop("random_rng_state"))
+        np.random.set_state(checkpoint.pop("numpy_rng_state"))
+
         if self._device == torch.device("cpu"):
-            torch.set_rng_state(checkpoint.pop("rng_state").to(self._device))
+            torch.set_rng_state(checkpoint.pop("torch_rng_state").to(self._device))
         else:
-            torch.cuda.set_rng_state(checkpoint.pop("rng_state").to("cpu"))
+            torch.cuda.set_rng_state(checkpoint.pop("torch_rng_state").to("cpu"))
 
         attr_names = [
             "dp_max_grad_norm",
@@ -447,4 +449,4 @@ def _update_from_checkpoint(self, path) -> dict:
         for attr in attr_names:
             setattr(self, attr, checkpoint.pop(attr))
 
-        return checkpoint
+        return

fedeca/algorithms/torch_webdisco_algo.py

@@ -3,7 +3,6 @@
 import os
 from copy import deepcopy
 from math import sqrt
-from pathlib import Path
 from typing import Any, List, Optional
 
 # hello
@@ -547,23 +546,18 @@ def _get_state_to_save(self) -> dict:
         checkpoint.update({"global_moments": self.global_moments})
         return checkpoint
 
-    def _update_from_checkpoint(self, path: Path) -> dict:
+    def _update_from_checkpoint(self, checkpoint: dict) -> None:
         """Load the local state from the checkpoint.
 
         Parameters
         ----------
-        path : pathlib.Path
-            Path where the checkpoint is saved
-
-        Returns
-        -------
-        dict
-            Checkpoint
+        checkpoint : dict
+            The checkpoint to load.
         """
-        checkpoint = super()._update_from_checkpoint(path=path)
+        super()._update_from_checkpoint(checkpoint=checkpoint)
         self.server_state = checkpoint.pop("server_state")
         self.global_moments = checkpoint.pop("global_moments")
-        return checkpoint
+        return
 
     def summary(self):
         """Summary of the class to be exposed in the experiment summary file.

fedeca/strategies/__init__.py

@@ -1,2 +1,3 @@
 """Init file for strategies."""
 from .webdisco import WebDisco
+from .bootstraper import make_bootstrap_metric_function, make_bootstrap_strategy