Why does the computer recommend rf4 ?

Why does computer recommend 19rf4 even though it does not leave me with any advantage later in the game and makes me a rook down ?, Please feel free to share your view as I am not great at chess. The png
to the game : [Event "Rated Classical game"]
[Site "lichess.org/eNbDtlWz"]
[Date "2018.04.04"]
[Round "-"]
[White "Bavan"]
[Black "killtheoppressor"]
[Result "1-0"]
[UTCDate "2018.04.04"]
[UTCTime "19:17:31"]
[WhiteElo "1360"]
[BlackElo "1354"]
[WhiteRatingDiff "+16"]
[BlackRatingDiff "-10"]
[Variant "Standard"]
[TimeControl "900+15"]
[ECO "A08"]
[Opening "King's Indian Attack: French Variation"]
[Termination "Normal"]
[Annotator "lichess.org"]

1. Nf3 { [%clk 0:15:00] } d5 { [%clk 0:15:00] } 2. g3 { [%clk 0:15:13] } c5 { [%clk 0:15:14] } 3. Bg2 { [%clk 0:15:19] } Nc6 { [%clk 0:15:26] } { A08 King's Indian Attack: French Variation } 4. e3 { [%clk 0:15:03] } Bf5 { [%clk 0:15:35] } 5. Nc3 { [%clk 0:15:09] } e6?! { (-0.30 → 0.26) Inaccuracy. Best move was Nf6. } { [%clk 0:15:48] } (5... Nf6 6. O-O e6 7. b3 Be7 8. Bb2 O-O 9. Ne2 h6 10. d4 cxd4 11. Nexd4 Nxd4 12. Bxd4) 6. O-O?! { (0.26 → -0.37) Inaccuracy. Best move was Nh4. } { [%clk 0:15:17] } (6. Nh4 Nh6 7. Nxf5 Nxf5 8. e4 dxe4 9. Nxe4 h5 10. h3 Qd4 11. O-O Rd8 12. c3 Qd7) 6... g6?! { (-0.37 → 0.41) Inaccuracy. Best move was Nf6. } { [%clk 0:15:54] } (6... Nf6 7. d3 Be7 8. b3 O-O 9. Bb2 d4 10. exd4 cxd4 11. Ne4 Qd7 12. Re1 Rfd8 13. Ned2) 7. d3 { [%clk 0:15:19] } Bg7? { (0.24 → 1.29) Mistake. Best move was Bg4. } { [%clk 0:16:06] } (7... Bg4 8. d4 cxd4 9. exd4 Bg7 10. Ne2 Nge7 11. h3 Bxf3 12. Bxf3 O-O 13. c3 Rc8 14. Nf4) 8. Nb1? { (1.29 → -0.32) Mistake. Best move was Nd2. } { [%clk 0:15:23] } (8. Nd2 e5 9. e4 dxe4 10. g4 Bc8 11. Ndxe4 Nge7 12. Bg5 O-O 13. Nf6+ Kh8 14. Qf3 h5) 8... Nf6 { [%clk 0:16:11] } 9. Qe1 { [%clk 0:15:33] } Ng4? { (-0.47 → 2.47) Mistake. Best move was O-O. } { [%clk 0:16:11] } (9... O-O 10. Nc3 d4 11. Na4 Nd7 12. exd4 cxd4 13. Nh4 Rc8 14. b4 b5 15. Nb2 Nce5 16. f4) 10. h3? { (2.47 → 0.68) Mistake. Best move was e4. } { [%clk 0:15:36] } (10. e4 O-O 11. exf5 exf5 12. Qd1 Re8 13. Nc3 d4 14. Ne2 Rc8 15. Nf4 Rc7 16. h3 Nf6) 10... Nge5 { [%clk 0:16:09] } 11. g4? { (0.86 → -0.76) Mistake. Best move was Nxe5. } { [%clk 0:15:39] } (11. Nxe5 Nxe5 12. Qe2 g5 13. e4 dxe4 14. dxe4 Bg6 15. Be3 Qe7 16. Qb5+ Qd7 17. Nc3 Qxb5) 11... Bxg4? { (-0.76 → 1.17) Mistake. Best move was Nxd3. } { [%clk 0:16:12] } (11... Nxd3) 12. hxg4?! { (1.17 → 0.24) Inaccuracy. Best move was Nxe5. } { [%clk 0:15:47] } (12. Nxe5 Nxe5 13. hxg4 Nxg4 14. f3 Ne5 15. Nc3 O-O 16. Bd2 h5 17. Ne2 Rc8 18. Bc3 b6) 12... Nxg4?! { (0.24 → 0.76) Inaccuracy. Best move was Nxf3+. } { [%clk 0:16:25] } (12... Nxf3+ 13. Bxf3 h5 14. g5 Qxg5+ 15. Bg2 h4 16. f4 Qe7 17. Bh3 g5 18. fxg5 Qxg5+ 19. Kh1) 13. Bh3 { [%clk 0:15:16] } f5 { [%clk 0:16:34] } 14. Nbd2 { [%clk 0:15:13] } Nce5? { (0.14 → 1.51) Mistake. Best move was Qd6. } { [%clk 0:16:21] } (14... Qd6) 15. Nxe5 { [%clk 0:15:02] } Bxe5?! { (1.51 → 2.05) Inaccuracy. Best move was Nxe5. } { [%clk 0:15:58] } (15... Nxe5 16. f4 Ng4 17. Nf3 O-O 18. Bd2 Qc7 19. Rb1 Rae8 20. Qg3 Nf6 21. Qh2 Nh5 22. Bg2) 16. Bxg4? { (2.05 → 0.72) Mistake. Best move was Nf3. } { [%clk 0:14:49] } (16. Nf3 Bc7 17. e4 dxe4 18. dxe4 Qd6 19. Qe2 O-O 20. e5 Nxe5 21. Nxe5 Qxe5 22. Qxe5 Bxe5) 16... fxg4 { [%clk 0:16:03] } 17. f3 { [%clk 0:14:39] } h5 { [%clk 0:15:51] } 18. fxg4? { (0.25 → -2.36) Mistake. Best move was e4. } { [%clk 0:14:31] } (18. e4 O-O 19. exd5 Bd4+ 20. Kh1 exd5 21. fxg4 Qd7 22. c3 Bg7 23. Qg3 hxg4 24. Rxf8+ Bxf8) 18... hxg4 { [%clk 0:15:50] } 19. Qf2? { (-2.36 → -5.20) Mistake. Best move was Rf4. } { [%clk 0:14:22] } (19. Rf4 Qg5 20. Qe2 O-O-O 21. Qg2 Bxf4 22. exf4 Qxf4 23. Nf1 Qf6 24. Be3 Rh3 25. Qxg4 Rdh8) 19... g3?? { (-5.20 → Mate in 1) Checkmate is now unavoidable. Best move was Qe7. } { [%clk 0:15:59] } (19... Qe7) 20. Qf7# { [%clk 0:14:13] } { White wins by checkmate. } 1-0

Hmm the game analysis is different from the one you get after you play and the the one you get after you import the game , as the first one says I made 0 blunders while it is not the case with the 2nd one

For easier viewing access, here's the game (blue link at the bottom left for Stockfish evaluations) :

All moves are losing, that one just puts up the most resistance that SF can see.

In human terms, the problem is that you're under massive attack on the kingside; you are, however, momentarily a piece up.

With that in mind, offering the exchange (a rook for a bishop/knight) to reduce the number of attacker (it cuts off the bishop) makes some sense.

If your opponent takes it, you actually aren't just down a rook; you have a knight and bishop for it (although you would still be worse, and your opponent could do better than taking it).

In short, you're just losing in that position, so SF is trying to delay a clear win for black as much as possible.

Cheers!

The computer wants the dark-square bishop's influence over the h-file nullified. The bishop, rook, and queen will (soon) be swarming your king. It is incredibly dangerous to allow all three pieces to remain involved in the attack, even if it means giving up material.

I would call 18. fxg4 a blunder; you should NOT have allowed black to take the h-file (with ...Qe7 and ...Qh7), from where your king will be vulnerable to black's attack. Closing off the kingside with 18. f4 was much better, although you would need to keep a careful watch over the now passed g- and h-pawns. (Better yet would be to blockade them; in the words of Aron Nimzowitsch, "The passed pawn is a criminal, who should be kept under lock and key. Mild measures, such as police surveillance, are not sufficient.")

I disagree with Stockfish either way; I believe white is slightly worse after 18. f4, but significantly so after 18. fxg4. There is certainly a computer defense, but black's position seems easier to play, with a clear plan to attack on the kingside and perhaps bring the king to the queenside. White's extra piece is almost entirely insignificant. The passed pawns alone are some compensation, but with an attack as well, I'd take black's position much sooner than I'd take white's.

What does Leela Zero think of this position?

I assume that given ample time, Stockfish evaluates that White is lost and all moves are equally bad.

@Toadofsky LC0 is mostly clueless in that position after black's 18th move.

It does realize black is better, but it doesn't realize the magnitude of the advantage.

I had it do a long-ish search (200,000 playouts), and to its credit, the winrate for white kept dropping, so maybe it's not quite as clueless as I indicated at first.

It does give most of the moves in the position a very similar eval (about 35% expected score), so in a way it realizes that everything sucks.

Having said that, it doesn't even try to put any resistance, either. It very quickly allows mate (19.e4 Qc7 20.d4); it has a ways to go :)