Pull out settings for model chat onboarding (#4274)

* Layer changes

* Documentation

parlai/crowdsourcing/tasks/model_chat/README.md

@@ -24,7 +24,9 @@ The following flags can be passed in to specify filepaths for overriding the tex
 
 ## Onboarding
 
-In `worlds.py`, modify `ModelChatOnboardWorld.check_onboarding_answers()` to change the worker selection criteria.
+Set the `"min_correct"`, `"max_incorrect"`, and `"max_failures_allowed"` fields in the JSON file passed to `mephisto.blueprint.onboard_task_data_path` in order to specify how many onboarding questions workers can pass/fail on while still passing onboarding, as well as how many times they are allowed to re-take the onboarding before being soft-blocked. (See `task_config/onboard_task_data.json` for an example.)
+
+You can further modify the worker selection criteria in `handleOnboardingSubmit` in `frontend/components/onboarding_components.jsx`.
 
 ## Human+model image chat
 

parlai/crowdsourcing/tasks/model_chat/constants.py

@@ -10,5 +10,3 @@
 
 ONBOARD_FAIL = '[ONBOARD_FAIL]'
 ONBOARD_SUCCESS = '[ONBOARD_SUCCESS]'
-
-ONBOARD_CONFIG = {'min_correct': 4, 'max_incorrect': 3}

parlai/crowdsourcing/tasks/model_chat/frontend/components/onboarding_components.jsx

@@ -9,9 +9,9 @@
 import React from "react";
 import { ErrorBoundary } from './error_boundary.jsx';
 import { Checkboxes } from './checkboxes.jsx';
-const ONBOARDING_MIN_CORRECT = 4;
-const ONBOARDING_MAX_INCORRECT = 3;
-const ONBOARDING_MAX_FAILURES_ALLOWED = 1;
+const DEFAULT_MIN_CORRECT = 4;
+const DEFAULT_MAX_INCORRECT = 3;
+const DEFAULT_MAX_FAILURES_ALLOWED = 1;
 var onboardingFailuresCount = 0;
 
 var renderOnboardingFail = function () {
@@ -35,7 +35,6 @@ function arraysEqual(_arr1, _arr2) {
 }
 
 var handleOnboardingSubmit = function ({ onboardingData, currentTurnAnnotations, onSubmit }) {
-    // OVERRIDE: Re-implement this to change onboarding success criteria
     console.log('handleOnboardingSubmit');
     var countCorrect = 0;
     var countIncorrect = 0;
@@ -60,10 +59,13 @@ var handleOnboardingSubmit = function ({ onboardingData, currentTurnAnnotations,
         }
     }
     console.log('correct: ' + countCorrect + ', incorrect: ' + countIncorrect);
-    if (countCorrect >= ONBOARDING_MIN_CORRECT && countIncorrect <= ONBOARDING_MAX_INCORRECT) {
+    const min_correct = onboardingData.hasOwnProperty("min_correct") ? onboardingData.min_correct : DEFAULT_MIN_CORRECT;
+    const max_incorrect = onboardingData.hasOwnProperty("max_incorrect") ? onboardingData.max_incorrect : DEFAULT_MAX_INCORRECT;
+    const max_failures_allowed = onboardingData.hasOwnProperty("max_failures_allowed") ? onboardingData.max_failures_allowed : DEFAULT_MAX_FAILURES_ALLOWED;
+    if (countCorrect >= min_correct && countIncorrect <= max_incorrect) {
         onSubmit({ annotations: currentTurnAnnotations, success: true });
     } else {
-        if (onboardingFailuresCount < ONBOARDING_MAX_FAILURES_ALLOWED) {
+        if (onboardingFailuresCount < max_failures_allowed) {
             onboardingFailuresCount += 1;
             alert('You did not label the sample conversation well enough. Please try one more time!');
         } else {
@@ -177,7 +179,7 @@ function OnboardingComponent({ onboardingData, annotationBuckets, annotationQues
                         className="button is-link btn-lg"
                         onClick={() => handleOnboardingSubmit({ 
                             onboardingData, 
-                            currentTurnAnnotations, 
+                            currentTurnAnnotations,
                             onSubmit,
                         })}
                     >

parlai/crowdsourcing/tasks/model_chat/task_config/onboard_task_data.json

@@ -85,5 +85,8 @@
                 ]
             }
         ]
-    ]
+    ],
+    "min_correct": 4,
+    "max_incorrect": 3,
+    "max_failures_allowed": 1
 }

parlai/crowdsourcing/tasks/model_chat/worlds.py

@@ -19,7 +19,6 @@
 from parlai.crowdsourcing.utils.worlds import CrowdOnboardWorld, CrowdTaskWorld
 from parlai.crowdsourcing.tasks.model_chat.bot_agent import TurkLikeAgent
 from parlai.crowdsourcing.tasks.model_chat.constants import (
-    ONBOARD_CONFIG,
     ONBOARD_FAIL,
     ONBOARD_SUCCESS,
 )
@@ -50,58 +49,14 @@ def __init__(self, opt, agent: "MephistoAgentWrapper"):
 
         self.skip_onboarding = opt['skip_onboarding']
 
-        self.min_correct = ONBOARD_CONFIG['min_correct']
-        self.max_incorrect = ONBOARD_CONFIG['max_incorrect']
         self.onboard_task_data = opt['onboard_task_data']
         self.status = 'DISCONNECT'
         self.onboard_statistics = opt['onboard_statistics']
         self.statistics_condition = opt['statistics_condition']
         self.max_onboard_time = opt['max_onboard_time']
         self.onboarding_qualification = opt['onboarding_qualification']
         self.worker_id = get_mturk_id_from_mephisto_wrapper(self.agent)
-
-    def has_same_answer(self, ans1, ans2):
-        if len(ans1) != len(ans2):
-            return False
-
-        ans1_sort = sorted(ans1)
-        ans2_sort = sorted(ans2)
-
-        for x in range(len(ans1_sort)):
-            if ans1_sort[x] != ans2_sort[x]:
-                return False
-        return True
-
-    def check_onboarding_answers(self, worker_answers) -> bool:
-        """
-        Calculate how many correct answers the user gave.
-
-        `worker_answers` is a list of dicts containing mappings between an annotation
-        value and whether it was selected for each bucket. We return a boolean as to
-        whether the worker passed or failed the task.
-        """
-        given_turns = self.onboard_task_data['dialog']
-        correct_answers = [t[1]['answers'] for t in given_turns]
-        number_correct = 0
-        number_incorrect = 0
-        for worker_answer, correct_answer in zip(worker_answers, correct_answers):
-            worker_only_selected = [
-                key for key, selected in worker_answer.items() if selected
-            ]
-            if self.has_same_answer(worker_only_selected, correct_answer):
-                number_correct += 1
-            else:
-                number_incorrect += 1
-
-        print(
-            f'Worker {self.worker_id} got {number_correct} annotations correct and {number_incorrect} incorrect in onboarding.'
-        )
-        if (
-            number_correct >= self.min_correct
-            and number_incorrect <= self.max_incorrect
-        ):
-            return True
-        return False
+        self.annotations = None
 
     def parley(self):
 
@@ -140,10 +95,11 @@ def _handle_act(self, act):
             print(f'{self.__class__.__name__}: {self.worker_id} had no data submitted')
             return ONBOARD_FAIL
 
-        worker_answers = act['task_data']['annotations']
+        self.annotations = act['task_data'].get('annotations')
+        print('Onboarding annotation results: ', self.annotations)
 
-        if self.check_onboarding_answers(worker_answers):
-            print(f'Worker {self.worker_id} successfully passed the onboard task.')
+        if act['task_data']['success']:
+            print(f'Worker {self.worker_id} successfully passed the onboarding task.')
 
             # This will end the onboarding and send them directly to the HIT
             self.episodeDone = True
@@ -164,6 +120,9 @@ def shutdown(self):
                     self.onboard_statistics[self.status] = 0
                 self.onboard_statistics[self.status] += 1
 
+    def get_custom_task_data(self):
+        return self.annotations
+
 
 class BaseModelChatWorld(CrowdTaskWorld, ABC):
     def __init__(self, opt, agent, bot):