ML, Deep Learning, XGBoost or CatBoost expert to solve a problem in data analysis
Budget: $250 – $750 USD
Need an expert to build a neural network or a decision tree able to solve a problem related to cryptography.
1) The Data
We will provide four listings containing nearly 300,000 lines of data each one, all of them similar to those presented in Examples 001 and 002, bellow.
The files are composed of letters A, C, T, G, B and (dots). Each of these letters corresponds to a "tag"
The (dots) corresponds to neutral tags (or simply, no tag). The letters ACTGB corresponds to a non-neutral tag (or simply different from "dot" tags).
The tags C, T, and G corresponds to small variations of tag A. They could even be made equal to tag A.
Just as a very simple example, lets consider:
If A = Antonio , then
C = Antonio wearing eyeglasses
T = Antonio wearing gloves
G = Antonio wearing a hat
C, T and G are small variations of the same A (Antonio).
Important: We can provide files containing more or less columns according to the needs of the expert who will build the Neural Network. The examples bellow contains 39 columns, which we consider suitable, but it can be easily changed under request.
Note: The files attached are composed of four different sets of data. To each of these files it will correspond a different neural network.
2) The problem
The aim of the project is to build a set of 4 Neural Networks able to identify those lines which could contain a tag B exactly at the central column (in the case of the examples bellow, the column number 20).
Any other B tag appearing outside of the central column can be ignored.
The only reference to identiry the correct lines (binary 1, tag B in central column) will be their A, C, T and G tags.
In the exemples bellow, the valid lines are marked with a binary 1 and the invalid lines are indicated with a binary 0.
3) The aim of the project
The aim is to identify AS MUCH AS POSSIBLE valid lines (tag B in central column) with the highest possible accuracy.
We are NOT interested to identify ALL binary 0 and all binary 1.
Identifying just a FEW valid lines (as much as possible binary 1 lines) will be enough, BUT THEY NEED TO BE INDENTIFiED WITH A HIGH ACCURACY.
Notes on former works on this project:
-- This problem has been already solved, in a former project, using the XGBoost software and the accuracy reached was 24% for those lines identified as valid (the lines containing a tag B in the central column). Unfortunatelly, this was a quite low accuracy and the amount of "FALSE' valid lines surpassed by far the amount of "REAL" valid lines. The aimed NN doesn't need to identify all REAL VALID lines but the amount of FALSE VALID lines needs to be quite smaller than these previously obtained.
-- The files are extremelly umbalanced, since the VALID LINES (binary 1) corresponds just to nearly 10% of all data.
EXAMPLES:
Note: All lines bellow have the same number of columns, the misalignment is due to the font.
Example 001 (training and validation file format):
.c..c....tt....a..t.t.ta.aba..aa..aa.at 0
..c....tt....a..t.t.ta.aba..aa..aa.at.. 0
....tt....a..t.t.ta.aba..aa..aa.at....a 0
..tt....a..t.t.ta.aba..aa..aa.at....a.. 1
.t.t.ta.aba..aa..aa.at....a....a...ag.. 0
.a...ag...ga.....atba...bt.a..a...t..ag 1
...ga.....atba...bt.a..a...t..agt....aa 0
..agt....aa.....b.a.gb..ttb.ba..t....tg 0
gb..ttb.ba..t....tgba.ag..a..aaa....a.a 1
..ttb.ba..t....tgba.ag..a..aaa....a.a.. 0
gba.ag..a..aaa....a.a..a.....aab.ga.... 0
Example 002 (working files - without any B tags)
.c..c....tt....a..t.t.ta.a.a..aa..aa.at
..c....tt....a..t.t.ta.a.a..aa..aa.at..
....tt....a..t.t.ta.a.a..aa..aa.at....a
..tt....a..t.t.ta.a.a..aa..aa.at....a.
.t.t.ta.a.a..aa..aa.at....a....a...ag..
.a...ag...ga.....at.a....t.a..a...t..ag
...ga.....at.a....t.a..a...t..agt....aa
..agt....aa.......a.g...tt...a..t....tg
g...tt...a..t....tg.a.ag..a..aaa....a.a
..tt...a..t....tg.a.ag..a..aaa....a.a..
g.a.ag..a..aaa....a.a..a.....aa..ga....
1) The Data
We will provide four listings containing nearly 300,000 lines of data each one, all of them similar to those presented in Examples 001 and 002, bellow.
The files are composed of letters A, C, T, G, B and (dots). Each of these letters corresponds to a "tag"
The (dots) corresponds to neutral tags (or simply, no tag). The letters ACTGB corresponds to a non-neutral tag (or simply different from "dot" tags).
The tags C, T, and G corresponds to small variations of tag A. They could even be made equal to tag A.
Just as a very simple example, lets consider:
If A = Antonio , then
C = Antonio wearing eyeglasses
T = Antonio wearing gloves
G = Antonio wearing a hat
C, T and G are small variations of the same A (Antonio).
Important: We can provide files containing more or less columns according to the needs of the expert who will build the Neural Network. The examples bellow contains 39 columns, which we consider suitable, but it can be easily changed under request.
Note: The files attached are composed of four different sets of data. To each of these files it will correspond a different neural network.
2) The problem
The aim of the project is to build a set of 4 Neural Networks able to identify those lines which could contain a tag B exactly at the central column (in the case of the examples bellow, the column number 20).
Any other B tag appearing outside of the central column can be ignored.
The only reference to identiry the correct lines (binary 1, tag B in central column) will be their A, C, T and G tags.
In the exemples bellow, the valid lines are marked with a binary 1 and the invalid lines are indicated with a binary 0.
3) The aim of the project
The aim is to identify AS MUCH AS POSSIBLE valid lines (tag B in central column) with the highest possible accuracy.
We are NOT interested to identify ALL binary 0 and all binary 1.
Identifying just a FEW valid lines (as much as possible binary 1 lines) will be enough, BUT THEY NEED TO BE INDENTIFiED WITH A HIGH ACCURACY.
Notes on former works on this project:
-- This problem has been already solved, in a former project, using the XGBoost software and the accuracy reached was 24% for those lines identified as valid (the lines containing a tag B in the central column). Unfortunatelly, this was a quite low accuracy and the amount of "FALSE' valid lines surpassed by far the amount of "REAL" valid lines. The aimed NN doesn't need to identify all REAL VALID lines but the amount of FALSE VALID lines needs to be quite smaller than these previously obtained.
-- The files are extremelly umbalanced, since the VALID LINES (binary 1) corresponds just to nearly 10% of all data.
EXAMPLES:
Note: All lines bellow have the same number of columns, the misalignment is due to the font.
Example 001 (training and validation file format):
.c..c....tt....a..t.t.ta.aba..aa..aa.at 0
..c....tt....a..t.t.ta.aba..aa..aa.at.. 0
....tt....a..t.t.ta.aba..aa..aa.at....a 0
..tt....a..t.t.ta.aba..aa..aa.at....a.. 1
.t.t.ta.aba..aa..aa.at....a....a...ag.. 0
.a...ag...ga.....atba...bt.a..a...t..ag 1
...ga.....atba...bt.a..a...t..agt....aa 0
..agt....aa.....b.a.gb..ttb.ba..t....tg 0
gb..ttb.ba..t....tgba.ag..a..aaa....a.a 1
..ttb.ba..t....tgba.ag..a..aaa....a.a.. 0
gba.ag..a..aaa....a.a..a.....aab.ga.... 0
Example 002 (working files - without any B tags)
.c..c....tt....a..t.t.ta.a.a..aa..aa.at
..c....tt....a..t.t.ta.a.a..aa..aa.at..
....tt....a..t.t.ta.a.a..aa..aa.at....a
..tt....a..t.t.ta.a.a..aa..aa.at....a.
.t.t.ta.a.a..aa..aa.at....a....a...ag..
.a...ag...ga.....at.a....t.a..a...t..ag
...ga.....at.a....t.a..a...t..agt....aa
..agt....aa.......a.g...tt...a..t....tg
g...tt...a..t....tg.a.ag..a..aaa....a.a
..tt...a..t....tg.a.ag..a..aaa....a.a..
g.a.ag..a..aaa....a.a..a.....aa..ga....