My latest creation during absent in EDMW

[Fidelis]

CNN to predict? Care to elaborate?

Sent from Chicken rice Phone using GAGT

There are a few parts:

* by plotting the occurrences as dots on a bitmap, TS has transformed the data representation from time domain to frequency domain

* true randomness in a man-made system is a myth because whether the numbers are picked using a mechanical system (such as balls) or digital system (such as pseudo-random number generator), the randomness is affected by imperfections in the balls and the source of the random number generator (biases)

* the goal of the exercise is to discover the imperfections (and accordingly biases) through analysis of the patterns, in this case extracting temporal-spatial discrepancies using neural networks (CNNs because the data have been represented graphically as bitmap but other neural nets can be used by just transforming the data domain accordingly)

* this is complicated by adjustments in the random generator methods used by the lottery organisers over the years because the underlying sources would have changed. These shifts can however be picked up through the output of the CNNs though if the analysis frames/windows are small enough

 

[Beetohven]

read liao dont understand. watch video even more blur. but going spool to buy 100 big 100 small nao.

well done ts

 

[Fidelis]

To add on, to detect imperfections/biases, we can work on training a neural network to recognise noise patterns generated by what we traditionally consider near-perfect entropy (e.g. weather patterns) and then it can be used to compute the delta between an arbitrary set of data inputs against "true entropy"

So primary goal is not to predict but to identify biases. The knowledge of underlying biases will then help create a predictor which can be used to validate our model

 

[fishbuff]

There are a few parts:

* by plotting the occurrences as dots on a bitmap, TS has transformed the data representation from time domain to frequency domain

* true randomness in a man-made system is a myth because whether the numbers are picked using a mechanical system (such as balls) or digital system (such as pseudo-random number generator), the randomness is affected by imperfections in the balls and the source of the random number generator (biases)

* the goal of the exercise is to discover the imperfections (and accordingly biases) through analysis of the patterns, in this case extracting temporal-spatial discrepancies using neural networks (CNNs because the data have been represented graphically as bitmap but other neural nets can be used by just transforming the data domain accordingly)

* this is complicated by adjustments in the random generator methods used by the lottery organisers over the years because the underlying sources would have changed. These shifts can however be picked up through the output of the CNNs though if the analysis frames/windows are small enough

i had a quick glance on some of the existing papers that talk about this. too hard for me to understand.

https://www.cs.montana.edu/sheppard/pubs/ijcnn-2019d.pdf
https://arxiv.org/pdf/1910.12708.pdf
https://arxiv.org/pdf/1803.03635.pdf

 

[fishbuff]

Great effort TS. Are you able to do a graphical view of the numbers?

100 x 100 bitmap, each pixel corresponds to one number

0000 corresponds to top-left pixel, 9999 corresponds to bottom-right pixel

Then allow user to select time frame

Within time frame, every time a number is picked, draw on pixel so that it gets bit darker. End of it, we have visual of the pattern for a time period

Bonus feature would be user can hover over a pixel to find out what number that is

Possible?

ah, so you propose to parse it into high dimension tensor, then analyze the pattern of each position over time. interesting... not sure where we can get the dataset for the past 3 years

saw this news before...

https://www.straitstimes.com/singap...ing-play-patterns-of-slot-machines-to-predict

"Two Russians jailed for recording play patterns of slot machines to predict next mass pay-out"

 

[Fidelis]

ah, so you propose to parse it into high dimension tensor, then analyze the pattern of each position over time. interesting... not sure where we can get the dataset for the past 3 years

saw this news before...

"Two Russians jailed for recording play patterns of slot machines to predict next mass pay-out"

Whether it is legal or not really depends on the jurisdiction and context of the analysis attempts

The slot machines were in a regulated environment and players are bound by rules that they agree to when they enter the premises. So their behavior can be clearly defined as cheating

Take a look at the article below:

https ://highline.huffingtonpost.com/articles/en/lotto-winners

But it’s also possible that math whizzes like Jerry Selbee are finding and exploiting flaws that lottery officials haven’t noticed yet. In 2011, Harper’s wrote about “The Luckiest Woman on Earth,” Joan Ginther, who has won multimillion-dollar jackpots in the Texas lottery four times. Her professional background as a PhD statistician raised suspicions that Ginther had discovered an anomaly in Texas’ system. In a similar vein, a Stanford- and MIT-trained statistician named Mohan Srivastava proved in 2003 that he could predict patterns in certain kinds of scratch-off tickets in Canada, guessing the correct numbers around 90 percent of the time.

 

[Fidelis]

To add on: the value of such a model can also come from being able to increase the entropy of systems requiring randomness (RNGs, etc) by funneling the output into GANs (Generative Adversarial Networks) so that successive generations will have fewer and fewer underlying biases that negatively impact the degree of entropy.

 

[fishbuff]

Whether it is legal or not really depends on the jurisdiction and context of the analysis attempts

The slot machines were in a regulated environment and players are bound by rules that they agree to when they enter the premises. So their behavior can be clearly defined as cheating

Take a look at the article below:

https ://highline.huffingtonpost.com/articles/en/lotto-winners

But it’s also possible that math whizzes like Jerry Selbee are finding and exploiting flaws that lottery officials haven’t noticed yet. In 2011, Harper’s wrote about “The Luckiest Woman on Earth,” Joan Ginther, who has won multimillion-dollar jackpots in the Texas lottery four times. Her professional background as a PhD statistician raised suspicions that Ginther had discovered an anomaly in Texas’ system. In a similar vein, a Stanford- and MIT-trained statistician named Mohan Srivastava proved in 2003 that he could predict patterns in certain kinds of scratch-off tickets in Canada, guessing the correct numbers around 90 percent of the time.

too hard. i will just stick with deep reinforcement learning and image segmentations

 

[Fidelis]

too hard. i will just stick with deep reinforcement learning and image segmentations

What are your thoughts on GPT-3? It shows promising signs that it could become the basis of the first ever AGI system?

https: //theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3

 

[fishbuff]

What are your thoughts on GPT-3? It shows promising signs that it could become the basis of the first ever AGI system?

https: //theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3

Thanks for introducing GPT-3 to me. i didnt know about that until you mention it. i have a read on Radford and Wu et al, 2019 paper later on.

could be. im not sure... according to my limited knowledge, i think GPT will only be a small service among the seas of ensembled models working together.

have you read deepmind's alpha-go zero papers? it is ensemble of MCTS, DRL, CNN... CNN for chess board recognition, mcts for decision selection, drl to explore known solution plus scoring. that is only one set just to play go. what if there are a complex stack of multiple levels of inter-connecting alphago zeros based model, each built and trained to serve different domain like sensory, detection, comprehension, text generation, image recognition etc... then a governing-coordinating one that gather all these input, plan the course of decision, then liaise with different other routine to work with the environment.

to me, trained model is one thing, but continuous update is another feature. making them to be more adaptive is my goal, so regardless of the base models for text/image recognition, anomaly detection, forecasting or optimization etc, it will need to recognise the ever-changing environment and prepare to explore for new solutions instead of relying on past historical data.

my 2cents
..........

 

[nubitol]

bump this thread up

 

[bwaysaigan]

Thanks for nubitol software

Managed to get 1 predictions spot on for consolations but i never buyed

 

[nubitol]

Thanks for nubitol software

Managed to get 1 predictions spot on for consolations but i never buyed

what kind of analysis u apply?

 

[nubitol]

lip lai lai lai

now version 2.1 liao

 

[Laneige]

Yr voice?????

 

[nubitol]

Yr voice?????

yeah my voice, why?

 

[busfreaks]

Tomorrow prediction no need use system. 72.51% huat

 

[bwaysaigan]

Tomorrow prediction no need use system. 72.51% huat

I got 2 digits same as yours. Also abcd

4 draws hope they ise back that same set balls.

 

[nubitol]

Tomorrow prediction no need use system. 72.51% huat

thanks, will buy but only $1

u bet so much, u very rich ah?

 

[busfreaks]

thanks, will buy but only $1

u bet so much, u very rich ah?

Money is everywhere just into whose pocket. Follow and u will be rewarded handsomely