1 (* Title: HOL/TPTP/mash_eval.ML
2 Author: Jasmin Blanchette, TU Muenchen
5 Evaluate proof suggestions from MaSh (Machine-learning for Sledgehammer).
10 type params = Sledgehammer_Provers.params
13 val MaSh_IsarN : string
14 val MaSh_ProverN : string
15 val MeSh_IsarN : string
16 val MeSh_ProverN : string
18 val evaluate_mash_suggestions :
19 Proof.context -> params -> int * int option -> string list -> string option
20 -> string -> string -> string -> string -> string -> string -> unit
23 structure MaSh_Eval : MASH_EVAL =
26 open Sledgehammer_Util
27 open Sledgehammer_Fact
28 open Sledgehammer_MePo
29 open Sledgehammer_MaSh
30 open Sledgehammer_Provers
31 open Sledgehammer_Isar
34 val MaSh_IsarN = "MaSh-Isar"
35 val MaSh_ProverN = "MaSh-Prover"
36 val MeSh_IsarN = "MeSh-Isar"
37 val MeSh_ProverN = "MeSh-Prover"
40 fun in_range (from, to) j =
41 j >= from andalso (to = NONE orelse j <= the to)
43 fun evaluate_mash_suggestions ctxt params range methods prob_dir_name
44 mepo_file_name mash_isar_file_name mash_prover_file_name
45 mesh_isar_file_name mesh_prover_file_name report_file_name =
47 val report_path = report_file_name |> Path.explode
48 val _ = File.write report_path ""
49 fun print s = File.append report_path (s ^ "\n")
50 val {provers, max_facts, slice, type_enc, lam_trans, timeout, ...} =
51 default_params ctxt []
52 val prover = hd provers
53 val slack_max_facts = generous_max_facts (the max_facts)
54 val lines_of = Path.explode #> try File.read_lines #> these
56 [mepo_file_name, mash_isar_file_name, mash_prover_file_name,
57 mesh_isar_file_name, mesh_prover_file_name]
58 val lines as [mepo_lines, mash_isar_lines, mash_prover_lines,
59 mesh_isar_lines, mesh_prover_lines] =
60 map lines_of file_names
61 val num_lines = fold (Integer.max o length) lines 0
62 fun pad lines = lines @ replicate (num_lines - length lines) ""
64 pad mepo_lines ~~ pad mash_isar_lines ~~ pad mash_prover_lines ~~
65 pad mesh_isar_lines ~~ pad mesh_prover_lines
66 val css = clasimpset_rule_table_of ctxt
67 val facts = all_facts ctxt true false Symtab.empty [] [] css
68 val name_tabs = build_name_tables nickname_of_thm facts
69 fun with_index facts s = (find_index (curry (op =) s) facts + 1, s)
70 fun index_str (j, s) = s ^ "@" ^ string_of_int j
71 val str_of_method = enclose " " ": "
72 fun str_of_result method facts ({outcome, run_time, used_facts, ...}
74 let val facts = facts |> map (fn ((name, _), _) => name ()) in
75 str_of_method method ^
76 (if is_none outcome then
77 "Success (" ^ ATP_Util.string_from_time run_time ^ "): " ^
78 (used_facts |> map (with_index facts o fst)
79 |> sort (int_ord o pairself fst)
81 |> space_implode " ") ^
82 (if length facts < the max_facts then
83 " (of " ^ string_of_int (length facts) ^ ")"
88 (facts |> take (the max_facts) |> tag_list 1
90 |> space_implode " "))
92 fun solve_goal (j, ((((mepo_line, mash_isar_line), mash_prover_line),
93 mesh_isar_line), mesh_prover_line)) =
94 if in_range range j then
96 val (name1, mepo_suggs) = extract_suggestions mepo_line
97 val (name2, mash_isar_suggs) = extract_suggestions mash_isar_line
98 val (name3, mash_prover_suggs) = extract_suggestions mash_prover_line
99 val (name4, mesh_isar_suggs) = extract_suggestions mesh_isar_line
100 val (name5, mesh_prover_suggs) = extract_suggestions mesh_prover_line
102 [name1, name2, name3, name4, name5]
103 |> filter (curry (op <>) "") |> distinct (op =)
104 handle General.Match => error "Input files out of sync."
106 case find_first (fn (_, th) => nickname_of_thm th = name) facts of
108 | NONE => error ("No fact called \"" ^ name ^ "\".")
109 val goal = goal_of_thm (Proof_Context.theory_of ctxt) th
110 val (_, hyp_ts, concl_t) = ATP_Util.strip_subgoal ctxt goal 1
111 val isar_deps = isar_dependencies_of name_tabs th
112 val facts = facts |> filter (fn (_, th') => thm_ord (th', th) = LESS)
113 fun get_facts [] compute = compute facts
114 | get_facts suggs _ = find_suggested_facts suggs facts
116 get_facts mepo_suggs (fn _ =>
117 mepo_suggested_facts ctxt params prover slack_max_facts NONE
119 |> weight_mepo_facts)
121 get_facts suggs (fn _ =>
122 find_mash_suggestions slack_max_facts suggs facts [] []
123 |> fst |> weight_mash_facts)
124 val mash_isar_facts = mash_of mash_isar_suggs
125 val mash_prover_facts = mash_of mash_prover_suggs
126 fun mess_of mash_facts =
127 [(mepo_weight, (mepo_facts, [])),
128 (mash_weight, (mash_facts, []))]
129 fun mesh_of suggs mash_facts =
130 get_facts suggs (fn _ =>
131 mesh_facts (Thm.eq_thm_prop o pairself snd) slack_max_facts
134 val mesh_isar_facts = mesh_of mesh_isar_suggs mash_isar_facts
135 val mesh_prover_facts = mesh_of mesh_prover_suggs mash_prover_facts
137 find_suggested_facts (map (rpair 1.0) isar_deps) facts
138 (* adapted from "mirabelle_sledgehammer.ML" *)
139 fun set_file_name method (SOME dir) =
142 "goal_" ^ string_of_int j ^ "__" ^ encode_str name ^ "__" ^
145 Config.put dest_dir dir
146 #> Config.put problem_prefix (prob_prefix ^ "__")
147 #> Config.put SMT_Config.debug_files (dir ^ "/" ^ prob_prefix)
149 | set_file_name _ NONE = I
150 fun prove method facts =
151 if not (member (op =) methods method) orelse
152 (null facts andalso method <> IsarN) then
153 (str_of_method method ^ "Skipped", 0)
156 fun nickify ((_, stature), th) =
157 ((K (encode_str (nickname_of_thm th)), stature), th)
160 |> map (fst #> nickify)
161 |> maybe_instantiate_inducts ctxt hyp_ts concl_t
162 |> take (the max_facts)
163 val ctxt = ctxt |> set_file_name method prob_dir_name
164 val res as {outcome, ...} =
165 run_prover_for_mash ctxt params prover facts goal
166 val ok = if is_none outcome then 1 else 0
167 in (str_of_result method facts res, ok) end
169 [fn () => prove MePoN mepo_facts,
170 fn () => prove MaSh_IsarN mash_isar_facts,
171 fn () => prove MaSh_ProverN mash_prover_facts,
172 fn () => prove MeSh_IsarN mesh_isar_facts,
173 fn () => prove MeSh_ProverN mesh_prover_facts,
174 fn () => prove IsarN isar_facts]
175 |> (* Par_List. *) map (fn f => f ())
177 "Goal " ^ string_of_int j ^ ": " ^ name :: map fst ress
178 |> cat_lines |> print;
183 fun total_of method ok n =
184 str_of_method method ^ string_of_int ok ^ " (" ^
185 Real.fmt (StringCvt.FIX (SOME 1))
186 (100.0 * Real.fromInt ok / Real.fromInt n) ^ "%)"
187 val inst_inducts = Config.get ctxt instantiate_inducts
189 ["prover = " ^ prover,
190 "max_facts = " ^ string_of_int (the max_facts),
191 "slice" |> not slice ? prefix "dont_",
192 "type_enc = " ^ the_default "smart" type_enc,
193 "lam_trans = " ^ the_default "smart" lam_trans,
194 "timeout = " ^ ATP_Util.string_from_time (the_default one_year timeout),
195 "instantiate_inducts" |> not inst_inducts ? prefix "dont_"]
196 val _ = print " * * *";
197 val _ = print ("Options: " ^ commas options);
198 val oks = Par_List.map solve_goal (tag_list 1 lines)
200 val [mepo_ok, mash_isar_ok, mash_prover_ok, mesh_isar_ok, mesh_prover_ok,
202 map Integer.sum (map_transpose I oks)
204 ["Successes (of " ^ string_of_int n ^ " goals)",
205 total_of MePoN mepo_ok n,
206 total_of MaSh_IsarN mash_isar_ok n,
207 total_of MaSh_ProverN mash_prover_ok n,
208 total_of MeSh_IsarN mesh_isar_ok n,
209 total_of MeSh_ProverN mesh_prover_ok n,
210 total_of IsarN isar_ok n]
211 |> cat_lines |> print