Ask Me Anything Free campaign and conversion analysis.

[Graybeard]

What the title says^

I have programmed an analysis program that can make a Linear Regression trends out of any x, y set of data

• impressions, ( campaign or siteid, date) x, y set of data
• ctr (to landing page), date) x, y set of data
• most other x, y set of data ---numbers are numbers

all data must be in a text file made with notepad, or exported in csv (from an advertising or network source), or from exel exported in csv.
fields (columns) separated by a ,

not sure what the minimum time should be but 60 - 90 days should be ok.

my objective it to test the accuracy we can make 2 projections

1. the first half of the data
2. the whole length of the data and you can see the trends.

This is a limited offer maybe I can do this 5 or 10 times to get a sample of its accuracy for CPA type stuff --networks are encouraged to participate
No disclosure of the location or exact nature is needed just two sets of numbers any x, y set of data you want me to test.

Example:

the x=value y=value i can convert (top right of chart)

mouse moved --you will have to mark off where after it is charted

this is an explanation (study course) describing liner regression principles.

Build and Interpret a Univariate Linear Regression Model

Build, interpret, and evaluate linear, logistic and polynomial regression models based on observations in your data.

openclassrooms.com

 

[Graybeard]

I worked up some different y data; FINRA short sales, Last Close and Daily trading volume --same program different data feed sources

this could be clicks, LP CTR and s2s sales conversions with some tweaks
GOLDEN OCEAN GROUP (GOGL-O)
Transportation / Freight & Logistics Svcs. / Marine Freight & Logistics

 

[LinkGenius]

This looks fun! I've played around with regression in the past.

Did you look to other regression models than linear? The last close data from your first post doesn't look linear.

Different Types of Regression Models

Regression models are prevalent in machine learning, and regression analysis is the most often used technique for problem solving.

www.analyticsvidhya.com

For me to understand what you are trying to do, could you tell a bit more about the meaning behind the values. Its not clear what you are trying to predict exactly, i.e. what are the values of last close (what does it mean if a data point has a last close of 130), volume etc. Similarly, you are making predictions based on time, is this relative to another point in time or just a date?

 

[Graybeard]

unix time is the x value =the date you do know what UNIX Time means? And Scientific notation right?

and the y values (vertical) is the price per share of GOGL a global bulk dry shipping business; grain, and other dry materials. Just an example. that is a linear regression chart. the scale is the price not 0 ..100

It's done with Python Linear Regression - statsmodels 0.14.0

Do you have at least 10,000 points of data? and some specific query to chart?

GOGL is 396,730,455 shares for the year data ; the contents of each trade is not known to me e.g.; 100 or 1000 share trades each.

in progres SQL: note the shorter time period NOT 1 yr

CREATE TEMP TABLE temp_symbol_data AS
SELECT last_day, lastsale
FROM finance.nasdaqdaily
WHERE symbol = 'GOGL'
AND last_day BETWEEN '2023-08-01'::date
AND '2023-11-10'::date;
 
SELECT 123
symbol |       slope        |     intercept      |      r_squared
--------+--------------------+--------------------+---------------------
GOGL   | 2816955.6883357265 | 1673835202.3837886 | 0.12641333051528245
(1 row)

Now what are you talking about? Your link is also Python Numpy

The last close data from your first post doesn't look linear.

#!/usr/bin/env python

import csv
import matplotlib.pyplot as plt
import numpy as np
from scipy.optimize import curve_fit
import datetime
from scipy.stats import linregress

while True:

 

[LinkGenius]

Thank for the explanation, I understand the examples much better now. I used regression for prediction mostly, so I thought polynomial or sinusoidal would fit the data of the very first graph better. But from an analytical point of view it makes more sense.

unix time is the x value =the date you do know what UNIX Time means? And Scientific notation right?

Yeah, should have looked that up.

You made me think this is actually a nice idea to have, so you can see how you are doing over time (visitors, commission, etc). So I asked my good friend chatGPT how to do this in Google Sheets. Turns out Google sheets has a trend line feature that does lineair (and polynomial and other) regression. Links are below if anyone wants to use it.

Google Sheet: Linear Regression

Lineair Regression

Blad1 Date,Value 07-01-2020,373.44 07-02-2020,303.92 07-04-2020,456.79 07-06-2020,920.94 07-07-2020,646.21 07-08-2020,1,332.73 07-09-2020,778.69 07-10-2020,893.94 07-11-2020,763.18 07-12-2020,986.17 07-01-2021,871.28 07-02-2021,1,483.47 07-03-2021,2,264.82 07-04-2021,3,381.74 07-05-2021,2,229.60...

docs.google.com

It's not as cool as doing it in python but I think it is easier for non programmers to use. Now I'm just thinking what kind of other data would be nice to use this for.

 

[Graybeard]

OK, that has been around for years in excel and I had been using Gnumeric scientific spreadsheets to do regression; linear fit, polynomial and exponential --manually for years.

That spreadsheet process takes time. I generate these graphs direct from csv data in about 2 to 3 seconds --but that is not the point.

Do you have at least 10,000 points of data? and some specific query to chart?

^^^ that is the point. I want datasets to prove, disprove, or hypothetically project trends on this "so called CPA data" with some degree of truth.
Thank for the visibility bump anyway

People ask about clicks rates CTR on creatives changing, conversions changing, etc all the time --trending this data may show something. Certainly more that pulling 'pat' answers out of the air. Or, what some hype blog or YouTube video says on the internet.

It could be this or that or maybe this --show me the data ...

 

[Graybeard]

this is what I am talking about --it's a work in progress ... so no one has any real CPA stats to test?

 

[Graybeard]

well anyway ...

it's all about predictable passive income no more work needed just need to monitor for any changes

 

[Graybeard]

well, moving on to a bigger pond with this

realtime no-pre market

 

[Graybeard]

Not done yet

Built a *Bloomberg Terminal*

Takes a minute or two to get all of the data.

The 4 x 4 charts to the right about 8- 12 seconds from local databases.
FINRA (top right) takes the time --it's a download each time.

I am trending --that is the red line

Now, if anyone had real web stats LOL anonymized is fine,

below is a mock up testing portfolio --see if this will really work

 

[Graybeard]

so no one had any real data to share? anonymized?

x,y Quadratic regression trending.

 

[Graybeard]

the video ...
250,000 pages aggregated in 2023 and analyzed for positive and negative news "affinity" by the mainstream media. Made with AI with my development planning --the charting --not finished yet but here is a peek:

The next video ads lolz ...