Redeploy carbon bot on Facebook

[samsucik]

I've added the 3 secrets that are needed to connect the bot to Facebook Messenger. Usually, the creds would go into credentials.yml as described here in the docs. Here, though, they're instead saved as Action secrets on this repo (as a repo admin, I can do this) and then plugged into the Helm chart.

[samsucik]

@kedz some more context is here.

[github-actions[bot]]

Intent Cross-Validation Results (5 folds)

class	support	f1-score	confused_with
macro avg	2517	0.72512739671103490	N/A
weighted avg	2517	0.80362214355672990	N/A
faq	643	0.78398791540785500	inform(22), inquire-ask_clarification-offsets(16)
inform	616	0.94060211554109040	faq(15), deny_flying(5)
affirm	255	0.86821705426356580	faq(7), estimate_emissions(5)
inquire-ask_clarification-offsets	124	0.70689655172413780	faq(30), why(3)
estimate_emissions	73	0.60759493670886080	faq(8), affirm(4)
deny	69	0.72131147540983610	faq(8), affirm(5)
insult	63	0.72058823529411780	faq(11), inform(1)
greet	63	0.87022900763358780	faq(4), inform_notunderstanding(1)
why	59	0.69421487603305790	faq(6), inquire-ask_clarification(5)
inform_notunderstanding	58	0.54901960784313730	faq(11), affirm(4)
farewell	57	0.83018867924528310	faq(5), insult(2)
thank	54	0.92592592592592590	greet(1), faq(1)
express_positive-emo	48	0.76000000000000000	SCENARIO(3), affirm(3)
vulgar	46	0.65753424657534230	faq(10), insult(10)
express_surprise	43	0.74418604651162780	faq(6), estimate_emissions(2)
express_uncertainty	43	0.73170731707317080	faq(6), greet(1)
inquire-ask_clarification	38	0.49315068493150680	faq(11), why(4)
buy_offsets	35	0.67532467532467530	faq(4), affirm(3)
how_calculated	29	0.80769230769230760	faq(5), estimate_emissions(3)
deny_flying	28	0.66666666666666650	faq(5), estimate_emissions(1)
express_negative-emo	25	0.62222222222222220	insult(2), inform_notunderstanding(2)
restart	18	0.88235294117647060	faq(2), affirm(1)
meta_inform_problem_bad-link	12	0.87999999999999990	faq(1)
SCENARIO	10	0.56000000000000000	faq(1), express_positive-emo(1)
help	8	0.42857142857142855	faq(3), inquire-ask_clarification(2)

Entity Cross-Validation Results (5 folds)

entity	support	f1-score	precision	recall
micro avg	926	0.8224500809498112	0.8220064724919094	0.8228941684665226
macro avg	926	0.7300728364723079	0.8033952924949558	0.7026989067354110
weighted avg	926	0.8210772453561229	0.8233933888922362	0.8228941684665226
city	384	0.8725361366622865	0.8806366047745358	0.8645833333333334
city.to	182	0.7903225806451614	0.7736842105263158	0.8076923076923077
city.from	149	0.7752442996742670	0.7531645569620253	0.7986577181208053
travel_flight_class	95	0.9285714285714285	0.9009900990099010	0.9578947368421052
iata	76	0.7083333333333334	0.7500000000000000	0.6710526315789473
iata.to	19	0.6341463414634148	0.5909090909090909	0.6842105263157895
iata.from	16	0.5600000000000000	0.7777777777777778	0.4375000000000000
number	5	0.5714285714285715	1.0000000000000000	0.4000000000000000

[samsucik]

@kedz thank you 🙂 I totally get that this is outside your usual scope. That's why I'm requesting reviews from you and Thomas -- this is a low-stakes situation (we can't break some important live deployment, so a bulletproof review from an expert isn't 100% needed) and I think it's useful for you to at least somewhat know what changes I'm making, so that the things I'm learning around CI/CD and FB integration don't stay completely within my own personal silo.

As for testing these particular changes before deploying them: I'm sure there would be a way but there's no obvious one. I guess if the bot was a high-stakes one, we'd set up different environments for staging deployments vs production deployments. But in order to test the connection to FB you'd sooner or later need to deploy the thing and just see if it works. I guess a bulletproof approach would then be to have two FB bots set up, one connected to the production deployment and one used for testing the staging deployments 🤔