Full Analyzing JZE's 29.43

[pauv]

Although there is already an analyzing video by Yowapon:
マインスイーパ上級世界記録29.43秒(Ze-En Ju)の徹底分析

Author Mao Igarashi（五十岚真央 from JPN）: msinfo ID=5403, high score=0.58+9.06+32.59, the former No.2 on World Ranking, twitter address=twitter.com/sweeper_yowapon, minesweeper blog=blog.livedoor.jp/yowapon/
And the translation:
日本玩家Yowapon对鞠泽恩29.43秒扫雷高级世界记录分析
Translator GONG Qiu-Yuan（龚秋源 from PRC）: saolei.wang ID=7939, high score=1.79+15.90+59.96, the author of the 《扫雷解局学》（Methodology of Minesweeperic Ressolvement），《扫雷学概论》（Minesweeperology）and 《基础扫雷教程》（Elementary Minesweeperics）
But the analyzing till now is still far from correct. To understand JZE's thought accurately, you need:
1. The understanding of NF skill of GWJ(GUO Wei-Jia, saolei ID=8356, msinfo ID=4425), which including Shao-Wu Zhang's op breaking skill (zhuanlan.zhihu.com/p/28703132), the click order selection and weijia's click technique, you've better master GWJ's full videos to be able to tell why he clicks here at any pause while watching his video at 1.00x speed. Like putianyi has point out, JZE is basically an FL-aid NFer.
2. The understanding of efficient FL skill of ZYY(ZHU Yao-Yu, saolei ID=14433, msinfo ID=6820), which including where to flag and where to chord. Any flag you put should has a clear target that which cells to open by chording and any chording set of cells should not be covered by each other. ZYY finished the highest exp IOE game in history(1.487) and not like Yi-Hai Lin, he is able to apply his high efficiency technique in fast solving so that he got 33s exp record. You've better read his full videos to master the various cases and practice to master his skill. JZE's thread about chording position by himself: minesweepergame.com/forum/viewtopic.php?f=24&t=1325
3. The understanding of the core thought of Advanced Minesweeper Experience on Smart Phone APPs by Wei-Jia Guo. Yowapon had found the core concept of "information first solving" when analyzing JZE's 29.43 but after reading GWJ's experience you will master the full theory and application of this concept. JZE is born to be a multi-thread player who is able to solve different parts of the minesweeper board at the same time, because he has done efficient practice on mobile phone to use both hands and control the 10+/s NF speed on beg level easily(although 6+ speed for each hand is enough but you will know the speed of brain is the true key factor if you try both hands ). While bringing the skill from mobile to PC, GWJ can do mid-range multi-thread NF operations, but JZE can carry out fluent long-range multi-thread flagging operations, which is the real difference between the best player and the 3rd best player.
4. The real cases of multi-thread flagging operations in JZE's full videos, you will feel it easier to master if you have watched some of JZE's streaming. JZE is not god, he does misses, wrong guessings, random decisions and anxious emotions.

In this thread, I will do long-term updating if I have time. I need time to do screenshots, compare with other similar cases in JZE's videos, ps pictures, record GIFs, imagine the possible strategy of other famous players and even review/fix my old posts. If I finish this thread I might analyze other WR by JZE(also step by step), and you are able to read JZE's other videos too.

--------------------------------------------
update: kiraa96(Michal Szoplik from POL, who achieved 0.45s non-lucky beginner WR, admin of minesweeper discord) has shared his analyzing about this game using English here and maybe further video reading from him will come.

[pauv]

Let's go!

JZE start.png (9.15 KiB) Viewed 865 times

The game start stage(stage 0): 0.00s - 0.21s
0.000 lc (7 10)
0.000 lr (7 10)
0.040 lc (7 35)
0.110 lr (6 104)
0.160 lc (6 142)
0.210 lr (9 155)
JZE uses 3 clicks at this stage, the 1st one hit the top left cell opens a big opening, the 2nd one hit the 2 on the left, the 3rd one hit the small opening on the left. No waste click.
At this stage, the main concern is look around the 1st opening to decide the next move. You have better clear the rest cell of the inner island, say, the green 1 cell, so that you needn't check that area in later process, and then clear the confusing cell between the openings like the green 2 cell, to use the cell as the base of pattern recognition for further sweeping. At the first glance, JZE notices several safe cells, the yellow ones, the green ones, and maybe the light green ones. Here, JZE has to make a decision about DIRECTION of the stage 1.

The principle of choosing direction is:
1. Which direction is more broader?
2. Which direction is next to the cursor?

The yellow cells influence the red region, while the green cells influence the blue region, it's obvious that the blue region fits the principle 1. Not mention the blue region fits the principle 2.

The stage 1: 0.52s - 1.29s
1st group
0.520 lc (5 88)
0.590 lr (7 87)
2nd group
0.760 lc (34 150)
0.920 lr (35 156)
1.020 lc (48 148)
1.090 lr (72 149)
1.140 lc (88 149)
1.180 lr (88 149)
1.240 lc (84 137) waste
1.290 lr (82 137)
JZE uses 5 clicks at this stage, hit the green 1~5 cells in order. The 5 clicks can divide into 2 operation groups, you will have muscle memory for each group if you are a pro NFer.
The most important cells here are the green 2 and green 3(I will identify the cell JZE is watching by red square in whole analyzing) to decide the next move. JZE gets the information of the cell 2 while hitting the cell 3 and gets the information of the cell 3 while hitting the cell 4, and also gets the safe information of the two light green cells for a backup solving strategy. However, the click hitting cell 4 breaks an opening, which makes the cell 5 a waste click. You may argue that hitting the cell 4 will be more efficient to break the opening, but you have no prior knowledge about the existence of that opening, for example if the number on green 2 is 3 and the number on green 3 is 2, you will lose many information in this case than JZE's choice, then you are forced to run out of your backup strategy in your next move or make stuck clicking on the cell 3 and the cell 5.

You may even grasp the concept of "视野 vision" here, for rookies they only have the 1x1 vision since they only see 1 cell at a time, so they are relying on safety chord to check what cells are next to a number. For some sub60 players like Amao(毛涤尘，Di-Chen MAO，saolei ID=19218), they have 2x3 vision, who are quite familiar with the 1-1 and 1-2 pattern. For sub50 player, they might have 5x5 vision, for example the red region. For sub40 player like me, I have 10x10 vision which is enough for the full grasp of both red region and blue region at a time. The key of a broad vision is not only the master of common patterns, but the accumulation of backup strategies, if 1 region has at least 1 strategy in it, I would control at least 2 backup strategies at a time for my decision, the more strategies you remember, the more selections you have, and the more efficient you are. For JZE, he is able to control several strategies at a time so that his vision can cover a half exp board. This is what such a level vision can do:

vision - GWJ 8.74on46.png (31.53 KiB) Viewed 748 times

This picture was drew by GWJ himself to explain what he saw at the very beginning of a int game. The red dots in the left image are all ensured safe at one glance, and a sweep route is shown at almost the same time in the right image. PS: Route deciding is not relying on the "safe" confirmation but the "solvable" confirmation, so the right image is even able to be realized faster than the left image.

vision.png (7.66 KiB) Viewed 863 times

A simple test of your vision: if you know the blue dot is safe at a glance, you have at least 7x7 vision. Amao who has 2x3 vision said 21-03-17: "True! I can only see 331 "
----------------------------------------------
update: Thanks to putianyi who provides me an example today showing the different vision cause different conclusion.
In his opinion, JZE only knows cell 1 and 5 in the first image are safe, the cell 234 are rely on the cell 1 clicking to react as safe, in another word, the cell 1 to 5 belongs to a continuous logic that the cell 1 is opened while the cell 2 is known as safe, the cell 2 is opened while the cell 345 are known as safe, which means JZE cannot look through the whole pattern in one glance.
However, such a process is too slow to sweep the board like JZE does, if you have the analyzing result that JZE's normal reaction time for new information is 0.22s. What putianyi does is actually dividing a big pattern into several small basic patterns, which is more friendly for rookies to understand but far away from JZE's view. Look at the difference between big pattern and small pattern:

vision - 2.png (7.33 KiB) Viewed 794 times

[putianyi888]

A simple test of your vision: if you know the blue dot is safe at a glance, you have at least 7x7 vision.

I can't solve it at one glance, but when I opened the opening at the top, I would keep in mind that the bottomleft of the 3 is a mine until I move to another region. A few seconds later, knowing that mine, I would click the blue dot without hesitation. I would call that pseudo-vision.

Memory plays a big role in fast solving. The more unflagged mines and unsolved parts you remember, the more choices you have.

[pauv]

stage 2.png (15.42 KiB) Viewed 842 times

After the stage 1, the whole board look like this. The green cells is 100% sure safe, the light green cell is possibly safe that needs further judgement, and on the right of the big opening there's an efficient area in green square(of course only a little impression). Keeping this "solvable impression" in mind as another backup strategy, JZE's main concentration is on the light green cell of the left side to make sure it's safe, and just like the latest 3 clicks in stage 1, there's an L-shape 3-Combo ended at the light green cell available, if the light green cell is proved to be mine in the first two clicks of 3-Combo, he can still click the left cell to finish the L-shape in different direction for keeping click rate, which is also the profit of reserving backup strategy in stage 1.
The stage 2: 1.51s - 2.68s
3rd group
1.510 lc (55 156)
1.590 lr (56 175)
1.640 lc (56 178)
1.700 lr (62 184)
1.750 lc (77 185)
1.920 lr (38 186)
1.970 lc (38 189)
2.130 lr (49 200)
2.170 lc (55 201) waste
2.220 lr (57 201)
2.280 lc (62 200) waste
2.350 lr (66 199)
4th group
2.510 lc (16 206)
2.590 lr (9 212)
2.640 lc (7 228)
2.680 lr (9 229)

JZE break op 1.png (3.76 KiB) Viewed 842 times

The light green cell is proved to be safe when the 1st click and is confirmed by the 1st click(red square), so it turns to green.
However, JZE realizes that the shape of the left is possible for breaking an opening, so he abandons the original plan and start the backup strategy immediately.
-----------------------------------
In fact, this weird shape shows again in this sup9 beg game:

JZE break op 1+.png (5.93 KiB) Viewed 842 times

Concentrating on the red square for further conclusion and click the cell below the 2.

JZE break op 2+.png (4.88 KiB) Viewed 842 times

Knowing the 4 green safe cells and click next to the 1(because of the red square 2, you know it's the L-shape 3-Combo), looking at the 1 just opened with the guessing of opening existence.

JZE break op 3+.png (5.13 KiB) Viewed 842 times

Knowing the yellow cell opening and click, looking at the rest corner to finish the game.
As a result, JZE cost 4 clicks for 1 opening! That's why this shape is weird.

[pauv]

JZE break op 2.png (3.83 KiB) Viewed 837 times

Back to the game. Similarly, JZE is going to click at the rest green cell (which also has the potential of opening breaking) while looking at the red square. When he finishes the clicking he knows the red square is 1 and the next two cells in horizontal direction is safe, to completely clear this area, he clicks the two cells and the green cell he abandoned before is the plan B in the case no opening is shown. The opening is confirmed after he clicks the light green cell. At the time of the latest click on the opening, JZE recognizes the whole opening shape and the island surrounding, abandons the plan B. 5 clicks are used on this opening, which makes JZE very frustrated.
Then, JZE clears the bottom left island and the 3 by NF, with the impression of an efficient solvable area on the right side. It's the same skill that the 3 is located when clearing the bottom left island, which is called 手眼分离(Separating the Cursor and the Sight), the basic of Multi-Thread Operation.
The whole process of stage 2 is below:

JZE 29.43on133 break op.gif (50.73 KiB) Viewed 837 times

As the preparation for the next stage, I'm going to introduce you the concept of "Shaping"(削皮).
You might have the experience of shaping apples, we call the skill of fast solving the islands touching an opening in order as "Shaping", which aimed to make the vision around the opening broader when solving the 3bv around the opening fast and efficient. Let's see Kamil's 7.03:

kamil shaping.png (388.49 KiB) Viewed 808 times

The good of Shaping: If you choose to clear the middle area completely now, you have the questionable light green cell in the left image which you have no information about, and it's difficult to recognize the pattern immediately, although you have the knowledge about the green cells are all safe. So Kamil follows the number order with fast left clickings, when it goes to the right image, the original difficult cell has enough information to be ensured as a safe cell. The shaping skill here is very efficient because all openings in the route is solved before, it provides you extra information of outer area helping you recognize pattern easier so that you can control more cells, and has more options to organize your clicking order.

The bad of Shaping: In the shaping process of the right image, since the information of bottom right corner is unknown, Kamil first clicking the cell 789 instead of cell 10(you are always able to solve cell 10 to have time to react about the information from cell 789), then an opening of yellow square is revealed at cell 8, making the cell 9 a waste click. The most efficient way to solve this area is to click the cell 9 directly and click cell 10 for reaction of the number shown on cell 9, then choose from plan A(if cell 9 is opening, click the bottom right corner if that's safe or goes to the bottom left corner by the 1-4 click order) and plan B(if cell 9 is not opening, click cell 7, cell 8 and the cell up to the cell 9 by order, then react to the information of the cell 7 and cell 8 for next move). The shaping skill here is not efficient because there is still an opening remaining, so waste clicks will be generated on the ops. If you have seen the exp videos of the Maestro of Shaping(I introduce his tips below), you will notice that he costs 6 or even 7 clicks on every opening, so that he has many 0.6+ thrp NF games(instance: 47.29, 107bv, thrp0.6407), the lowest thrp avg!

The-Importance-of-Opening-Breaking.gif (14.12 KiB) Viewed 680 times

JZE's FL Shaping: if it's JZE solving this area, he would put a flag up to the yellow square and chord for the four green cells(using 1.5-click skill), then click cell 10 while reacting to the number just shown in the yellow square. This is much more stable on efficiency for both opening case and island case than NF Shaping!

As conclusion, NF Shaping is the dual skill of Opening Breaking, it's efficient and easy to use while there are no opening, however, it's inefficient while there are unsolved openings. Beautiful instances of NF shaping from top NFers:
Instance of GWJ's NF shaping: 8.74s on 46 with 5.26 3bv/s
Instance of JZE's NF shaping: 9.13s on 57 with 6.24 3bv/s
Beautiful instances of FL shaping: all JZE's FL videos and ZYY's FL videos.

The premise using Shaping:
1. Master the skill of Opening Breaking.
2. Only shaping where is not opening.
3. Needs tons of information for decision-making of directions in short time.
4. Having the ability to change direction and jump over mines when clicking in the case of high speed moving.

Tips of Shaping practice from Hao-Yu Li(李浩宇, 削皮宗师 - the Maestro of Shaping, saolei ID 12689):
1. Shaping based on your vision and continuity of pattern reaction.
2. Your vision seems like 2x7 or even 2x9 when shaping.
3. There's nothing difficult, just be very familiar with easy patterns.
4. Recognize all mines around 1 and 2 aside your route while shaping upwards, then downwards, vice versa.

[pauv]

shaping 0.png (13.95 KiB) Viewed 785 times

The stage 3: 2.90s-5.21s
5th group
2.900 lc (90 200)
2.960 lr (88 199)
3.130 lc (152 183) waste
3.210 lr (159 180)
3.280 lc (177 172)
3.350 lr (177 170)
6th group
3.460 rc (196 166) 1.5click-r
3.500 lc (198 164) 1.5click-d1
3.560 lr (197 151)
3.590 lc (203 150) 1.5click-d2
3.650 lr (208 150)
3.690 rr (204 157) 1.5click-end
7th group
3.800 rc (167 166) 1.5click-r
3.830 lc (167 166) 1.5click-d1
3.890 lr (177 168)
4.070 lc (188 175) 1.5click-d2 waste
4.140 lr (182 180)
4.240 lc (162 185) 1.5click-d3
4.300 lr (165 185)
4.330 rr (176 190) 1.5click-end
8th group
4.450 rc (199 200) 1.5click-r
4.500 lc (203 193) 1.5click-d1
4.570 lr (214 182)
4.630 lc (213 187) 1.5click-d2 waste
4.700 lr (212 201)
4.740 rr (216 206) 1.5click-end
9th group
4.890 rc (247 200) 1.5click-r
4.920 lc (250 198) 1.5click-d1
4.970 lr (260 197)
5.020 lc (262 204) 1.5click-d2 waste
5.070 lr (263 215)
5.210 rr (259 216) 1.5click-end
At the end of stage 2, JZE solves the cell 1 and 2 in order, then goes for the cell 345 to reach the efficient yellow square found in the beginning of stage 2. The clicking on the cell 4 and 5 are necessary for the information of the light green area b. If it wasn't a misclick on the cell 4 JZE would chord on the cell 4 after this image since it's 1. JZE then starts 1.5-click skill, flags the cell 6, chord for the green c cells, and chord for the light green d cells while receiving the information from the cell 5 as a 2. (I use red line for a group of continuous flag and chord operations, the red line always starts at a mine and ends at the latest chording cell.)

choding 0.png (14.56 KiB) Viewed 785 times

It's efficient to use 1.5-click skill here, at least for the green cells. (The 1.5-click here is equals to press right button -> 3-Combo -> release right button.) Flag the 2 with 3 chordings around the mine so that the deep blue area is covered, and the orange cell is confirmed as mine while the 2nd wasted chording. Then flag+2 chord for the blue area, and then the purple area. As shown in the picture, the flag & chording direction is decided by this principle: Each area influenced by each chord group should not be covered by each other.

shaping2multi-thread.png (88.89 KiB) Viewed 785 times

The stage 4: 5.34s-9.60s
10th group
5.340 lc (208 216)
5.430 lr (199 216)
5.500 lc (200 199) waste
5.560 lr (204 194)
5.700 lc (155 214)
5.790 lr (151 219)
11th group
5.900 lc (121 200)
6.000 lr (118 198)
6.060 lc (112 198)
6.130 lr (108 199)
12th group
6.170 lc (106 212) 3-Combo
6.330 lr (133 230)
6.410 lc (146 234)
6.470 lr (152 234)
6.520 lc (159 234)
6.580 lr (166 234)
13th group
6.680 lc (127 219)
6.750 lr (121 218)
6.780 lc (108 218)
6.850 lr (103 218)
14th group
6.980 lc (108 229)
7.060 lr (105 234)
7.100 lc (115 234)
7.160 lr (120 234)
15th group
7.300 lc (105 249)
7.390 lr (104 249)
7.450 lc (114 249)
7.490 lr (116 249)
14th group
7.670 lc (167 246) 3-Combo
7.750 lr (167 246)
7.800 lc (175 246)
7.850 lr (182 245)
7.900 lc (183 233)
7.960 lr (180 231)
8.020 lc (180 231) waste
8.070 lr (180 233)
15th group
8.220 lc (211 233)
8.300 lr (211 233)
8.410 lc (227 216)
8.470 lr (230 214)
8.550 lc (230 225)
8.630 lr (230 231)
8.670 lc (229 235)
8.720 lr (229 245)
16th group
8.830 lc (247 244)
8.920 lr (247 244)
9.020 lc (270 230)
9.100 lr (278 229)
9.140 lc (285 228) waste
9.600 lr (267 195)
Stage 4 is the 1st highlight in this game.
Here goes the image ①. Group 10 and Group 11 belongs to this image, after the latest chording of stage 3, JZE feels very awkward that there's no 1 for chording and the only thing he can do is to click the cell a. And then a waste click was made for decision making and releasing his click rate, he thinks that the route of the red arrow is the best choice for later solving. However, the yellow area is very long and has no information about, which is also possible to have an unknown opening while clicking the yellow cells, and is dangerous for a possible 50/50. For this reason, after the cell b clicking he should concentrate on the information under cell b, a deal of profit would be made if a long opening was to be found when clicking the cell c-d-e-f.
So, in the image ②, regarding e and f as backup strategy after the a to d shaping(f is also uncertain square), he click 1 to 3 and then 4 to 5(Group 12 and 13), for cell 5 is confirmed safe when 1 to 3 clicking.

Now, please check the image ③, the vertical light green cells are safe because of the 3-4 pattern, the horizontal green cells are safe because of the mine wall to the left side, and the horizontal light green cells might be safe(this is a concept of "预判(predict)", guess the most possible case, the best case or the worst case for the next 2 steps). To clear this area completely, it's time to restore the vertical light green cells as backup strategy, clicking the horizontal green cells for the information of horizontal light green cells, so he solves the two green cells from the left to the right, then luckily the light green cells are both confirmed to be safe. So in the image ④ it's possible to click 1 to 5 continuously, which is not optimal order for you have better click the yellow cell first for possible opening, 4-5-3(GWJ would click like 4-5-1-2-3 if 4 is not opening) will be better than 3-4-5.

As we observe, in the image ⑤ which is a combination of the image ① to ④, there're 2 threads, the red arrow thread that solved yellow area and the green arrow thread that solved blue area. The red thread divided/solved the yellow area by 4 layers, the green thread divided/solved blue area by 3 layers. The 1st blue layer and the 1st yellow layer are solved in the same process of shaping, but the 2nd and the 3rd yellow layer are solved while the green thread is idle. As a summary, some tips about Multi-Thread Operations are below:

1. Multi-Thread Operations organizing the whole board (or the whole region you are solving) as several logical area.
2. Multi-Thread Operations is rely on player's memory about other logical area, separating the cursor and the sight, and the skill of shaping. The broader your vision is, the more threads you can choose from.
3. Multi-Thread Operations solving each logical area in one thread, some threads are idle when some threads are active.
4. An active thread can become idle and become active again when needed.
5. Multi-Thread Operations is the realize of information solving, the only reason that a thread is active is the current layer of the logical area which the thread is working on containing the most important information that player needs.
6. Layers from different logical area can be solved in the same shaping process, while it's not a must to involve different logical area in one shaping process.
7. Shaping nearby logical area back and forth is the fastest way, which fully covers the famous Kamil's skill 1, since it's more efficient, faster and less rely on luck (betting for possible 1).