fix: update to reflect the fact that coin's learn does not return a p…

…rediction (#4621) * fix: coin's learn does not return a prediction, update to reflect The prediction it currently returns essentially represents what the prediction would have been post learn and therefore does not match the contract of learn_returns_prediction, * update test --------- Co-authored-by: Alexey Taymanov <[email protected]>
VowpalWabbit · Aug 31, 2023 · bea6816 · bea6816 · reportgambling · Sep 1, 2023
1 parent 66d5572
commit bea6816
Show file tree

Hide file tree

Showing 3 changed files with 20 additions and 20 deletions.
diff --git a/test/pred-sets/ref/ftrl_coin.predict b/test/pred-sets/ref/ftrl_coin.predict
@@ -1,12 +1,12 @@
 1
 0.361325
-0.143347
+0.143348
 0
 0.067602
 0.934070
 0.160364
 0
-0.576177
+0.576178
 1
 0.068731
 0
@@ -38,17 +38,17 @@
 0
 1
 1
-0.118569
+0.118570
 0.791537
 0
 0
 0.001875
 0
-0.179304
+0.179305
 0.141783
 0.575180
 0.236666
-0.860866
+0.860867
 1
 0
 0
@@ -95,7 +95,7 @@
 1
 0
 0.683733
-0.138811
+0.138812
 0.555727
 0.665974
 0.212627
@@ -147,8 +147,8 @@
 0.616288
 1
 0
-0.636642
-0.694423
+0.636643
+0.694424
 0.643842
 0.647201
 0.032352
@@ -163,13 +163,13 @@
 0
 1
 0.250740
-0.752841
+0.752842
 0.756894
 0.112263
 0.912395
 0.563831
 0.576449
-0.142849
+0.142850
 0
 0.390992
 0
@@ -188,7 +188,7 @@
 0.207041
 0.094113
 0.555100
-0.612930
+0.612931
 0.217877
 0.319611
 0.119250

diff --git a/test/train-sets/ref/ftrl_coin.stderr b/test/train-sets/ref/ftrl_coin.stderr
@@ -16,13 +16,13 @@ Output pred = SCALAR
 average  since         example        example        current        current  current
 loss     last          counter         weight          label        predict features
 1.000000 1.000000            1            1.0         1.0000         0.0000       51
-0.503420 0.006841            2            2.0         0.0000         0.0827      104
-0.253545 0.003669            4            4.0         0.0000         0.0000      135
-0.250782 0.248019            8            8.0         0.0000         0.0259      146
-0.278089 0.305396           16           16.0         1.0000         0.1788       24
-0.292183 0.306278           32           32.0         0.0000         0.3479       32
-0.263581 0.234979           64           64.0         0.0000         0.0007       61
-0.233750 0.203920          128          128.0         1.0000         0.7720      106
+0.503420 0.006841            2            2.0         0.0000         0.0000      104
+0.253545 0.003669            4            4.0         0.0000         0.0291      135
+0.250782 0.248019            8            8.0         0.0000         0.0620      146
+0.278089 0.305396           16           16.0         1.0000         0.1084       24
+0.292183 0.306278           32           32.0         0.0000         0.1627       32
+0.263581 0.234979           64           64.0         0.0000         0.0626       61
+0.233750 0.203920          128          128.0         1.0000         0.7347      106
 
 finished run
 number of examples = 200

diff --git a/vowpalwabbit/core/src/reductions/ftrl.cc b/vowpalwabbit/core/src/reductions/ftrl.cc
@@ -458,7 +458,7 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::ftrl_setup(VW::setup_base_
         all.output_config.audit || all.output_config.hash_inv ? learn_coin_betting<true> : learn_coin_betting<false>;
     all.weights.stride_shift(3);  // NOTE: for more parameter storage
     b->ftrl_size = 6;
-    learn_returns_prediction = true;
+    learn_returns_prediction = false;
   }
 
   b->data.ftrl_alpha = b->ftrl_alpha;
@@ -498,4 +498,4 @@ std::shared_ptr<VW::LEARNER::learner> VW::reductions::ftrl_setup(VW::setup_base_
                .set_print_update(VW::details::print_update_simple_label<ftrl>)
                .build();
   return l;
-}
+}