Statistics & Data Science related scripts
Code by Dries Bultynck
Language: English
Questions, bugs or requests: dries@driesbultynck.be or dries@parafix.io
Script: get an overview of all the accounts, profiles and views with their filters, custom dimensions & custom metrics
To do: needs some work
Basic template for machine learning applications and needed/possible steps. Clean sheet.
Script: Predict needed reach based on desired clicks and real historic reach data from past reach campaigns as potential reach indicated by Facebook isn't trust worthy and doesn't allow brands to really invest in market share as they aim to do (TV budget -> GRP build-up vs online budget -> reach build-up).
Model: link clicks models is accurate - first screening: avg. of 6x budget needed to possibly reach the effective potential reach
To do: done
Technique: Linear regression
Script: Check correlation + cross lag impact, decompose trends & seasonality, forecast traffic & show causal impact
Model: accuracy to check
To do: done
Script: Impact of type of day on Revenue - days like: day of week, holiday, closing days, ...
Model: accuracy to check
To do: done
Script: Market basket analysis based on combined page visits per session (optional: per user-id or client-id)
Model: accuracy to check
To do: done
Technique: Market basket analysis
Script: Correlation & prediction of outcome based on weather data & Google Analytics goals
Model: accuracy to check - outliers detection needs to be better -> capping datapoints
To do: on-going
Script: Vizualisation of channels & share of total based on sessions, transactions, revenue and more
Model: no model
To do: forecasting scenarios with conversion rates etc. or attribution depending on focus
Script: Basic channel attribution
Model: no model
To do: order & export needed to tie to budgets
Technique: Markov
Script: Basic density & facet mapping of data
Model: no model
To do: clean up, more examples, better viz
Technique: none
Script: Export & graph of campaign stats per day, week, month, quarter, year ... based on labeling with custom reporting by metrics liek conversion rates, spend/return, weighted values, etc.
Model: no model
To do: graphs
Technique: none
Script: Explore data quality & distribution
Model: no model
To do: test
Technique: none
Script: Graph of distribution by pageDepth (converters vs non-converters) to initiate behavioural segmentation
Model: no model
To do: turn into segments
Technique: k-means? or weighted return on pageDepth?
Script: Graph cumul impact by campaign over time
Model: no model
To do: /
Technique: time normalization
Script: Graph cumul impact by landing page over time
Model: no model
Focus: /
Technique: time normalization
Script: GA & Screaming frog merged data with rapid clustering for top problem detection
Model: k-means on sessions & url part detection per cluster for priority definition
To do: topic modeling & NLP for clustering on entity and context -> labeling & automatic outlier exclusion
Technique: k-means
Script: Google Adwords screening
Model:
To do:
Technique:
- Colour codes for graphs based upon book of leila
- Types of graphs based upon book of leila
- Make markdown of graphs or specific scripts for reporting in pdf
- Anomaly detection on sessions or days of year over one year to make campaigns accountable or detected spam issues
- Make several machine learning template according to supervised or unsupervised techniques
- K-means clustering on user-id, session-id, urls, ...
- Neural network training with tensorflow + output in xls or api?
- Pipeline management from input, model, output into working environment/database with R or Python
- Make functions for packages to load, handling data, etc.
- Make the general analytical marketeer and/or its reporting obsolete for GA