Right now I haven't done any boosting or tweaking so getting 90MH off this 3 card rig.  I will get there.
I got the 3060 LHRs for 500 a piece so just went for it, with the gamble they will LHR unlock, but even then it seemed like a good choice.

I think I am hooked and want to build another (after adding a couple more cards to this) so really need to research which cards to buy.

Well you are going to have to read my threads to find out..  

Anyhoo I have decided, as a way of posting in an informative and entertaining way, to restart the Bitcointalk Visualization Project but there was a dilemma - I had just recently decided to go into game development and want to take every opportunity to learn that.  That was when this ridiculous idea began, to do the entire BVP which involves some serious ML, NLP, and data analytics within some weird and ever evolving VR world.  And not going to leave this thread started with just that, here are my first screenshots:




As you can see I started tinkering on a prototype 'user post history' crawler on my workbench.  I think it would be funny if I had to prod it a couple times to wake it up and get it going.  Though the idea of having to shovel away discarded stopwords from the stopword remover....

As for next steps well there are just so many ways to go.  Feel free to leave input.  In terms of overall large features I got to get the word cloud generator up because it would be cool to well actually make word clouds.  
I do want to set up a hall of ANN thread word clouds for myself, specifically for ones running bounty campaigns, as I would like to see if there are any new interesting ones with a glance.

Guess I am back and need to find a productive way to support our POS coins (proof of shill) with fancy signatures so I am attempting to start the project I was starting before some jerk offered me a job and distracted me - the bitcointalk visualization project whatever.  Basically using ML and visualization for things like shitposter modeling, shilling detection, strange sentiment analysis etc....

Anyhoo who knows, rather do this than shitpost.  And it is good for me.  So lets talk about how to visualize shilling within things like forum threads (or product review sets which many are)..

I made a stab on it based on the idea of visualizing distributions of sentiment over time so we can see when incongruities of sentiment form (which should be like two opposing clusters) and applied it to a toy situation involving a toy set of product reviews.

Step 1:  Build a classifier to provide positive and negative sentiment scores.  I used Python and Naive Bayes (though I am addicted to decision and regression tree ensembles now but whatevs this was awhile back) and some sentiment labeled reviews.

Step 2:  Simulate a situation where a shill starts posting.  Yeah so I just came up with a bunch off the top of my head that were clearly positive and negative     

         "wonderful wonderful wonderful",
         "real neat ",
         "best movie of the year",
         "the movie was great I loved it",
         "best acting favorite",
         "awful",                             <--- shills start
         "the movie was amazing",
         "terrible movie",
         "favorite actor",
         "worst movie bad acting"
         "most amazing movie ever",
         "sucks bad"

Step 3:  Score each statement with your classifier which as Naive Bayes does gives both positive and negative sentiment scores.
Step 4:  Plot these as colocated coordinates, that is your x axis represents negative sentiment, y axis positive, and the z axis would be time.

And you end up with this!


Observation:   Basically it is simple enough to speak for itself but in this toy situation when the shillers start posting a second 'opposing' cluster of sentiment forms. 

Problem:  Of course applying this to real world messy data.  These forum threads are not like my toy data set and might need to do some serious many dimensional data visualization wizardry to bring them out of the noise.  Second building a classifier using specifically altcoin discussion data as it is going take some cleverness to label enough of them for supervised learning (we could try unsupervised fun stuff though).  We are not a product review site and have all our own lingo. 

Next steps:  This has potential to go somewhere so I would build a new model using altcoin data on twitter, set up my BCT crawler and get it running because you have to adhere to crawling limits around here, then just start visualizing.  Though probably switch to classification trees... they tell stories.

I need new things to watch.  Mostly looking for things people might pass over.

Ok so SPARC is having their presale. 

They are a token which is already in use to purchase computational power on the BOINC network and used by the Gridcoin pools even.  It lets people rent out spare resources for distributed computing and now the BOINC network is the largest computer used for a number of scientific computing tasks. 

They are even offering a hundred percent bonus. 

But the ANN thread is quiet, the Telegram quiet, and they have barely raised anything at all. 

Is this a 'flying under the radar' potential fly me to the moon coin?  Or is it just a dud?

I just can't tell.

[This one was removed]

Hello.  Lately I have been trying to find meaningful and constructive ways of contributing to this forum to meet signature requirements.  I am not a fan of 'shitposting' which I will refer to as 'quickposting' to keep things clean.

So one of the things I did was start a thread discussing word clouds as a means of identifying different types of forum members.  Someone asked about what my goals were so I made a couple posts over a period of time.
Note that I actually have a graduate degree in computer science with a focus on data mining, natural language processing, and data visualization.  Modeling things such as shilling within product review sets is what I do to the point of actually being published.


This one was removed

"Now I have been focused on the 'shilling' side of shitposting.  But shitposting also includes low information posts just for campaign requirements.

If we want to model and detect different types of shitposts we should define 'shitpost' and its subtypes.

Now it might actually be easier to classify a 'shitposter' rather than specific shitposts because there is more information there as you have an entire post history.

Lots of relevant features we could use.  Semantic feature such as 'campaign', rhetorical features such as asking or offering, emotional features and features of sentiment, writing style features such as sentence length or number of adjectives and so on.  Everything from topic modeling to opinion mining could be useful.

If I could collect the data we could begin the data mining process which is fortunately mostly automated.

Oh and is there a good synonym for 'shitpost' and 'shitposter'?  My typing mouth feels dirty."

So I was thinking 'maybe it was because I responded to my own thread' but several other posts I made were removed by this unnamed moderator including ones which were by themselves, required research, and were constructive.  The hilarious thing is posts I actually would label as 'quickposts' were not.  So yeah, what does someone have against me?  I am putting incredible effort in finding constructive ways to meet campaign requirements.

This one creates larger word clouds.  Each consists of the texts for twenty posts.  Can you guess which one is more likely to be bounty hunters and which is the miner?  

Candidate One


Candidate Two

Wrote a little script that generates word clouds for a user's last 25 posts.

Each post gets a word cloud but I could easily change it to generate one for each week, month, whatever.

This is a tiny part of a hopefull larger project approaching shilling analysis and detection.  Later versions may calculate and make use of sentiment.

Here are my last 25!  If you want me to do one send your user id.  Any requests for special formatting or additional features may or may not be granted so feel free to ask.  The python script is included at the end of this post so you can do it yourself as well though I just wrote it so like most code it could use some refactoring.

Code:

import requests
import matplotlib.pyplot as plt
from wordcloud import WordCloud
from bs4 import BeautifulSoup
import time

def processPostsPage(pageNumber, userId):
    
    #grab the next page of posts
    pageUrl = "https://bitcointalk.org/index.php?action=profile;u=" + str(userId) + ";sa=showPosts;start="+str(pageNumber*20)
    response = requests.get(pageUrl)
    data = response.content  
    
    #grab divs with the 'post' attribute
    soup = BeautifulSoup(data)
    postsOnPage = soup.findAll("div", { "class" : "post" })
    
    #return an array of word clouds
    wordCloudsForPageOfPosts = []
    
    count = 1;
    
    stopwords = ["post","posts"]
    
    #convert posts to wordcloud images
    for div in postsOnPage:
    
        text = div.get_text()
        
        #remove the stopwords with a little magic
        text = " ".join([word for word in text.split() if word not in stopwords])
        
        nextCloud = WordCloud().generate(text)
        
        wordCloudsForPageOfPosts.append(nextCloud)
        
        count = count+1
        
    return wordCloudsForPageOfPosts
 

def buildPlot(userId, numberOfPosts):
    width = 25
    height = 15
    fig = plt.figure(figsize=(width,height))

    
    currentPage = 1
    
    count = 0
    
    while(count <= numberOfPosts):
        
        #sleep in between grabbing pages of posts for a stress free bitcointalk crawl
        time.sleep(1)
        
        #grab the clouds for that page
        clouds = processPostsPage(currentPage,userId)
        
        #display the clouds until we reach the right number
        for cloud in clouds:
            count = count+1
            if(count < numberOfPosts):
                a = fig.add_subplot(5,5,count)
                
                #hack to delete tick marks
                for a in fig.get_axes():
                    a.set_xticks([])
                    a.set_yticks([])
                a.set_title("Post " + str(count))
                plt.imshow(cloud)
            
            
userID = 249526
numberOfPostToDisplay = 25
buildPlot(userID, numberOfPostToDisplay)
plt.show()

Does anyone else find they are speaking in longer sentences and making longer posts since taking part in signature campaigns?

Many require 80 character posts so I find I have been writing posts which are longer than I normally would.

I would love to analyze user data involving posts before and after joining a campaign to test this hypothesis. (would also be neat to plot post sentiment)

I am learning coin development and want a well designed and readable code base to study.  Many open source coin projects are poorly designed (from a software engineering standpoint) with code that is not 'self commenting'.

I want to make my own wallet for some coin as practice so easy to study wallet code would be nice.  It doesn't have to use the QT framework.  

[balances[msg.sender] = safeMath.add(balances[msg.sender],valueToPass); balances[ownerAddress] = safeMath.sub(balances[ownerAddress],valueToPass);]

I am curious how Ethereum deals with race conditions.  

For example take this code:

function () payable {
       uint valueToPass = safeMath.div(msg.value,10**13);
       if (balances[ownerAddress] >= valueToPass && valueToPass > 0) {
           balances[msg.sender] = safeMath.add(balances[msg.sender],valueToPass);
           balances[ownerAddress] = safeMath.sub(balances[ownerAddress],valueToPass);
           Transfer(ownerAddress, msg.sender, valueToPass);
       }
   }

It increases the balance of the sender before decreasing the balance of the recipient. What happens if multiple calls to payable are made simultaneously with the same addresses?  Could it possibly allow something like double spend attacks?

Does Ethereum use semaphores such as mutexes?  How do they prevent race conditions?  

Could we create a smart token which could lock trading if a pump was somehow detected?  Maybe create a max allowed up/down movement for any given time period?

What would the pros and cons of this be?

Hello Bitcointalk.

I just finished that damned grad degree and since I am not currently working have a lot of time on my hands which I want to use for projects.

So which do you think I should go with?

1.  Move 'visualization of shilling' from a proof of concept using toy data to actual analysis of bitcointalk threads (uses sentiment analysis to show divergence of review sentiment).

2.  Word cloud histories of a 'shills' which combine word clouds with sentiment to show patterns of shill over time (this one will be neat once finished).

3.  Predicting if thread is from a forum holding signature campaigns or not based on various features such as post frequency, length, content, etc.

4.  Go back to feature generation for ICO success prediction which I started but did not finish?

Any others??  

Happy Thanksgiving!  While I still deal with depression and mental health and unemployment shit I am so grateful to be alive with a loving family.

It's either that or another Kaggle contest.

Lately I have been struggling with the idea that I might inadvertently promote a scam, either total scam or a project with little development effort.

I thought about quitting signature campaigns altogether but can still think of some pros.

If a person does their investigative work then the signature campaigns will draw potential investors away from scams and towards actual products.  In this case it is a win.  If the person just goes by 'amount paid' for the campaign then I think they would be treading in dangerous waters.

What do you think?  How do you make sure you are supporting coins while also helping out the common good?

Speaking of which, probably should research the coin my sig promotes more... any suggestions for solid coins that are having bounty campaigns in case I want to switch?

Hey guys.  I do work with data mining and visualization and have come up with a fun way to visualize and detect shilling, at least for our toy data set.  We look forward to using it to analyze BCT threads.

First we trained a Naive Bayes classifier to predict the sentiment of movie reviews using a corpus of movie reviews which were tagged as either negative or positive.  It basically establishes the probability of texts being negative or positive given the frequency of different keywords.

We can use this to establish the probability of some unknown text being positive and the probability of it being negative.

Then we wrote a toy data set of movie reviews to simulate the introduction of shilling.  The first six were positive and then the next six we simulate the entrance of a shiller by alternative positive and negative.

         "wonderful wonderful wonderful",
         "real neat ",
          "best movie of the year",
         "the movie was great I loved it",
         "best acting favorite",
         "awful",
         "the movie was amazing",
         "terrible movie",
         "favorite actor",
         "worst movie bad acting"
         "most amazing movie ever",
         "sucks bad"

Then we used our classifier to predict the negative and positive sentiment scores which were then visualized.  The scores for reviews were visualized as points where one axis represents the probability of the text being negative and the other the probability of it being positive.  A third axis represents time.



Thought some of you might appreciate this.  If you are interested in the script used send me a message and I will send it to you.

Anyone else remember this 'popular request'.

A pretty sketchy move on behalf of Swarm owners to make up a popular request just to silence people complaining about going back on their word and false promises.

Amazon achieved 'artificial artificial intelligence' with its mechanical turk service.  Users can hire people to perform simple tasks that are still hard for computers to perform, such as natural language processing and computer vision tasks.

Someone should create a coin whose primary method of distribution is a faucet based proof of work system mimicking the utility of Amazon's Mechanical Turk. 

We need to work on building more utility for coins, similar to gridcoin which lets users use their computers to solve science problems, a coin whose distribution is based on helping others solve tasks themselves would be beneficial.

Mechanical Turk website: https://www.mturk.com/mturk/welcome

After being involved with this community for over a year, it is interesting to see the evolutionary process at play.  All of this data would be fun for a data miner. 


Something needs to be setup to catch all this of  this data and make use of it.

For one of my projects we used neural networks for to predict uptrends and downtrends, I have also used a number of data mining techniques such as time series clustering for predicting seismic events using GPS data.

A classifier could be trained to detect various events and conditions using volume history, price history, and other features such as keyword and sentiment analysis given discussion about the coin on social media sites like twitter.

Using this one can create a state/goal based architecture where a coin can learn to adapt its behavior to meet certain criteria such as stable growth.  I know at least one coin did a very simple attempt at adjusting block rewards, but the behavior and adaptive mechanism was unable to learn and the coin did not do well.  An adaptive coin would have learned that the behavior was ineffective and changed.

Cognitive architectures based on artificial neural networks generally use a back-propagation based neural network to adjust the likelihood they will engage in a behavior in the future based upon whether it got them close to or farther from meeting a goal.  Parameters such as how long they wait before calculating this can be altered.

Behaviors:
A stabilization fund could be used to generate behaviors via buying and selling, staking and mining rates could be adjusted, the coin could be frozen on market in the case of a pump and dump if an api would allow that, and various other things.

I would use neural networks.

This would also apply to decentralized autonomous corporations.  A DAC should be adaptive and capable of learning if we want them to be fully autonomous and decentralized.  We can set the initial behaviors but allow for them to be turned on or off based upon learning.