News to date
- 2024/Nov: Added DoubleML via EconML.
This popular model has been requested several times.
In addition, we improved data file reading, which should be more reliable now given non-ASCII characters and unusual date-time formatting.
If you have problems getting Causal Wizard to read your CSV or Excel files, ensure you have a header row and
read our dataset guide. Feel free to contact us for help.
- 2024/Oct: Causal Wizard has been growing in popularity and now has over 700 unique users per week. On average, each website visitor reads 10 pages per visit, so we are pleased our educational mission is going well!
- 2024/Sep: Minor updates to improve page load speed. Some major new features and upgrades are on the way.
- 2024/May: Our first birthday! Causal Wizard has been helping people to adopt Causality in their
research for one year.
Since we do not advertise, it's been a slow start, but we're pleased that over 4,300 unique users have used
our tools and read our articles.
On average, each user reads 12 pages of content and half our visitors return for more.
- 2024/Feb: Edit-Study and Results pages now have numbered titles, making it easier for you to manage,
find and compare results when you have may tabs open.
- 2024/Jan: CausalWizard developers have been contributing to the Open-Source DoWhy Python library, which
provides the key Causal ML
methods we use. We aim to push the features we really need to deliver CausalWizard back into DoWhy, so
everyone can benefit from them.
- 2024/Jan: Significantly enhanced the Exploratory
Data Analysis (EDA)
tools built into CausalWizard. It's vital that users can explore and check their data,
visualise individual variable value distributions, and examine bivariate relationships (between pairs of
variables).
CausalWizard now includes Histograms, Contour and Scatter plots for numerical association,
multiple Violin and Box plots for mixed categorical / numerical association and
Heatmaps for categorical / categorical analysis.
- 2023/Dec: Added feature-importance analysis to
provide insights into the modelled behaviour and
effects of covariates on outcome.
- 2023/Dec: Option to define a held-out test set in your data and perform Machine-Learning style
predictive validation techniques on it, including classification metrics for categorical outcomes and
R-squared analysis of continuous outcomes.
- 2023/Nov: Added Positivity analysis (key feature),
including Propensity distribution and
covariate-balance plots.
This feature is enabled whenever a propensity
score
method is used. It validates that your Control and Treated groups cover the same range of values for
all input variables, which is important for unbiased results.
- 2023/Nov: Added contingency table to results (a summary of all treatment and outcome combinations in
data).
- 2023/Sep: Added support for Categorical data (major update).
- 2023/Aug: AI/ML project designer tool added:
- 2023/Aug: Speaking about Python Causality tools at the 2023 PyCon
conference in Adelaide, Australia.
- 2023/Aug: Speaking about Causality and Causal Wizard at the MLAI
meetup in Melbourne, Australia.
- 2023/Jul: Enabled drawing of Causal Diagrams without uploading data, by popular request.
- 2023/Jul: Counterfactual outcomes added.
- 2023/Jun: Help article database and videos added.
- 2023/Jun: Began working with two client organisations to help them model causal effects in their
data.
- 2023/May: Basic Exploratory Data Analysis (EDA) features added to Datasets.
- 2023/May: Causal Wizard launched.
High priority new features / changes
Features with strikethrough style have already been added.
Predictive performance validation: Allow a percentage of the data to be held out as validation or
test set, and measure predictive performance on it.
Feature-importance analysis: Provides insights about the modelled behaviour of variables used in
regression models.
Automatic normalization / standardization: Would allow better feature-importance interpretation
with
regression models.
Support for categorical variables, String data types, and ability to mark a Variable as
categorical
even if
content is numerical.
Positivity (overlap) test: Check whether data satisfies criteria, possibly filter samples which
fail
the test.
Add support for additional estimation methods from EconML.
- Augmented dataset download: This is our top priority. We will enable you do download the fully preprocessed version of your data we use in the models, including individual model predictions, counterfactual predictions, and categorically encoded variables. This feature will provide greater transparency and utility, enabling you to analyse your data and results offline.
- Enhanced Wizard features: More sophisticated and comprehensive recommendation system.
- How-to articles: Examples of Causal Inference applied to various industry problems.
Lower-priority features / changes
Question: Do you prefer interactive plots on the result page, or rendered images? Tell us!
- Enhance page load speed (smaller JS files).
- PDF downloads with selectable text rather than image.
- Export Causal Diagrams in various machine-readable formats.