jordantaylor.io
About this site...
Wecome to my site that I use to highlight professional interests. From time to time I attempt to use open-source libraries to solve what I believe are unique problems within the maritime industry. I am partnered with Joe Gaudiano, Bsc., MBA, an experienced east coast ship operator. You can contact me at tayljordan(at)gmail.com or LinkedIn.
Applications
TC VALIDATE is a platform for estimating the voyage revenue of tank vessels. Using vessel identification systems and proprietary routing software the application will will be able to identify spot vessels that will arrive within laytime cancelling as well as calculate vessel revenue in real-time. The platform will provide clarity with regard to fair market value and allow commercial managers to make clean, evidence-based decisions.
LINK TO APPLICATION
The project has been developed as a minimal viable application by Joe Gaudiano and Jordan Taylor, both experienced and driven maritime professionals. Dr. Jenya Antonova, a senior economist in the healthcare industry, is advising on machine learning modeling components.
Indication Natural Language Toolkit
Back-end application for parsing indications. Textual emails can be broken down into constituent parts for application to a database. The data can then be used for further analysis. MVP project is about 50% complete.
Papers
Test APIs
Attempts to find optimal - and less optimal - bulk liquid packing solutions among different sized tanks. The outcomes will create some draft stows that can be used as a basis for further analysis, particularly when considering temperature, adjacency and technical considerations.
Extraction of computer-readable dates and times from laytime statements. This API will serve as the first stage of a tool that can assist with understanding demurrage exposure using natural language processing.
Using 10-year International Maritime Organization casualty data in conjuction with regression analysis using gradient tree boosting, this API will attempt to assess a vessel's risk of casualty based on location and type of activity the vessel is engaged in.
Summary
A brute-force bin packing API.
Used for placing different size liquids into different sized containers. This question may seem simplistic however it is a classically difficult problem to solve.
In:
-
Tank volumes
-
Product volumes
Out:
-
Possible stowage combinations
Calling the API
There is a package written in python that can call the API here.
Simplified Example
Arguments:
-
tanksVolValue = [20, 20, 10, 5]
-
prodsVolValue = [20, 25]
Tuning arguments:
-
s = 3
-
r = 1
-
Depth = 10
-
Processes = 1
Result:
Considering tuning arguments above there are 18 total combinations expressed as a list. Each element in the list contains a variation of the following:
[[20, 25], [7003], {'7003': 10}, [[7002], [7004, 7001]]
Breakdown:
[20, 25] = The product volumes, sorted
[7003] = Remaining tank that is empty and available. The 700 in '700X' has no meaning, it was an identification holdover from early development.
{'7003' : 10} = remaining empty tank(s) and their volumes
[[7002], [7004, 7001]] = The stow. i.e. the first product 20 will fit into the second tank and the second product will fit into the first and last tank.
So - to recap - if given,
7001 = 30
7002 = 20
7003 = 10
7004 = 5
then the first volume (20) will be stowed in 7002 and the second volume (25) will be stowed in 7004 and 7001. 7003 is an available tank.
Front end screenshots utilizing API
Example of self-stowing chemical tanker application,
Summary
Dates and times are extracted from vessel statement of facts and converted to a computer readable format. This API will be the first step in automated processing of laytime statements. The examples below are the results of a classification algorithm made specifically for laytime analysis.
In:
-
A vessel statement of fact in textual format
Out:
-
Vessel event dates and times converted to a computer readable format
-
Event data that will be used for further analysis
Calling the API
API development is ongoing.
Example of Date and Time Extraction
In:
FEBRUARY 22nd 2017
14:30 END OF SEA PASSAGE
15:18 ANCHORED AT JOSE BAY
FEBRUARY 23rd 2017
00:01 NOTICE OF READINESS TENDERED
10:36 AGENT/PORT AUTHORITIES ON BOARD
11:06 FREE PRATIQUE GRANTED
13:40 PORT SECURITY ON BOARD
Out:
0 2017-02-22 14:30:00 end of sea passage
1 2017-02-22 15:18:00 anchored at jose bay
2 2017-02-23 00:01:00 notice of readiness tendered
3 2017-02-23 10:36:00 agentport authorities on board
4 2017-02-23 11:06:00 free pratique granted
5 2017-02-23 13:40:00 port security on board
Example of Natural Language Processing using Levenshtein Distance
Below are the inputs and results from statement of fact-derived raw data. The number preceding the output is the confidence of the processor in obtaining the final result. Note that broken English, misspellings and stemmed phrases are able to be handled. It should be noted that common NLP packages, such as Stanford CoreNLP, are unable to handle maritime technical phrases - thus the reason for employing this specific method. A production model will likely require a technical corpus.
In:
eosp in posn n w
nor tendered
pilot onboard
cahnge of pilot
let go port anchor down vessel anchored
Out:
089 end of sea passage
100 notice of readiness tendered
100 pilot onboard
100 pilot change
051 anchored
Final Outcome
Using above processing it will be relatively easy to map statements in order to generate commercial outcomes.
Some - but not all - byproducts of mapping:
-
Demurrage exposure
-
Understanding the implications of multiple tendering of NORs
-
Unusual delays to include weather and bunker delays
-
Regression analysis
-
Performance indicators
-
Using machine learning to understand potential commercial outcome of vessel.
Limitations
Lack of a technical library.
Front end screenshots utilizing API
Example of raw statement-of-fact processing using the API,
