# impactchart
[](https://firstdonoharm.dev/version/3/0/cl-eco-extr-ffd-law-mil-sv.html)
[](https://vengroff.github.io/impactchart/)
[](https://vengroff.github.io/impactchart/)
`impactchart` is a Python library for generating impact charts.
## What are Impact Charts?
(See also the [ACM FAccT paper](https://facctconference.org/static/papers24/facct24-80.pdf) for more background and details.)
Impact charts make it easy to take a data set and visualize the impact of one variable
on another in ways that techniques like scatter plots and linear regression can't,
especially when there are other variables involved.
For example, consider this impact chart, which looks at the impact of
`x_3` on `y` in a particular data set:
The green dots represent our best estimate of the
impact. The grey dots around them represent the estimate of the impact based on many
(in this case 50) different machine learning models. When they are close to the green
dots, as on the left side of the chart it means there is strong agreement among the
models as to the impact. When they are farther apart, as on the left side of the chart,
there is less agreement.
The general shape of the curve of green dots, and the fact that the gray dots remain
rather tightly grouped around it, suggest that the impact of `x_3` on `y` is very limited
when `x_3` is negative. But as it becomes increasingly positive, it's impact grows more
and more rapidly. It might even be exponential \[Spoiler alert: this is synthetic data and
the impact is exponential.].
Now let's compare what we see in the impact chart to a traditional scatter plot, which
is a popular way of taking a first look at the relationship between variables.
Using the same data set, here is a scatter plot of `y` vs. `x_3`:
In the scatter plot, we also added regression curves for linear regression and
quadratic regression, but they did not tell us much. The reason is that
in the data set we are looking at, `x_3` isn't the only feature that impacts
`y`. There are other `x_i` whose cumulative impact hides that of `x_3`.
The impact chart, on the other hand, we see the impact of `x_3` on `y` independent
of the effect of any other `x_i`.
The reason impact charts like the one we are looking at are so powerful is that they
very clealy and directly show us the impact of one feature in a data set on a target
of interest.
Impact charts can find impact even though, unlike parametric regression techniques,
they don't have any _a priori_ knowledge of the shape of the impacts they are looking for.
For example, in the data set we have been looking at, there is another feature `x_2` whose
impact on `y` is sinusoidal. And the impact chart for it shows this clearly
In general, no matter what the shape of the impact of one variable on another,
whether it is polynomial, exponential, logarithmic, polylogarithmic, has seasonality
or other repetitive structure, if the impact is there you are going
to get a good visualization with an impact chart.
And what if there is no impact? What if the two variables are not correlated at all.
We can see an example of that case in the following impact chart:
In this case, the impact chart is essentially flat at zero. There is a little bit
of random noise resulting from the way we created and trained the underlying machine
learning models, but the fact that there `x_4` had negligible impact on `y`, at least
relative to the impacts `x_2` and `x_3` had, is quite clear.
## What do Impact Charts on Real Data Look Like?
The data set we looked at above to introduce impact charts is synthetic. We deliberately
constructed it so some of the variables had known impacts and others had none. It was
part of a test to show that the ideas behind impact charts worked and that the code
that we wrote to implement them was working.
But of course the real reason for impact charts is to look for impacts that we did not
deliberately put there for the code to find. So what do impact charts look like on real
data? One of the first data sets we tried it on consists of data about the median income of renters,
the racial and ethnic make up of the population of renters, and the rate of eviction at the
census tract level in hundreds of communities around the United States. We made impact charts
at the county level to show the impact of income, race, and ethnicity on eviction.
Here is an example:
This chart looks at the impact of the median household income of renters (at the census tract level)
on the eviction rate in DeKalb County, Georgia in years between 2009 and 2018. The impact is measured
in units of eviction cases filed per 100 renters.
We might hypothesize that the lowest income renters
would have the highest eviction rates. But this impact chart tells us otherwise. In very low income
tracts, there is an impact around -5, meaning that these tracts have an eviction rate five points
lower than they otherwise would. Why would that be? In some cases, those at the very bottom of the income
scale live in public housing or voucher-supported housing (commonly known as Section-8). In these
kind of settings, they have more protection against eviction than renters do on the open market.
Between $25,000 and $40,000, the impact is almost entirely above zero. These tracts experience higher
rates of eviction because of their income level. This is consistent with a population of people
who have enough income to pay rent under normal circumstances but not enough to absorb the kind
of shock (a major car repair or medical bill) that those with higher income might have the savings
or credit to survive without being evicted. Above $50,000 in household income, the curve is mostly
flat at zero impact. This doesn't mean none of these people ever get evicted. It just means that
it is not their income that is driving the eviction.
But between $50,00 and $100,000, the green
dots are bifircated into a zero impact group and a group around -3 to -4. Usually this means
there is some other variable that is causing the difference between the two, but it was not
present in our data set, so the models did the best they could in assigning that impact. If we could
add tha right variable to our data set, the effect would go away.
In just one chart, we can start to build a very interesting picture of the dynamics of the social and
economic processes behind the data. And we can compare the impact of variables. Without going into
all of the details we discuss
[elsewhere](https://datapinions.com/the-impact-of-demographics-and-income-on-eviction-rates/)
here is another impact chart from the same data set. But instead of looking at the impact of
median renter income, we look at the impact of the fraction of renters in a census tract who
identify as Black alone. The chart is:
Census tracts where very few Black renters live have a negative impact, meaning their eviction
rates are lower. The dats set include, income, whose impact we already looked at, so this
impact chart shows the impact that can be attributed to race independent of income, even if
the two are correlated to some degree. Note also that the two impact charts share the same
vertical scale, so that the magnitudes of the impacts can be compared.
If you would like to see more impact charts generated from this data set, please visit
[http://evl.datapinions.com/](http://evl.datapinions.com/).
## How Do I Make an Impact Chart
The first step is to install the `impactchart` code in a virtual environment using
```shell
pip install impactchart
```
From there, the simplest way to make your first impact chart is to
replicate the code that generated the first impact charts we saw above, using
synthetic data. It is as follows:
```python
from impactchart.model import XGBoostImpactModel
from impactchartdemo.dataset import synth1
# Generate the data set:
N = 200
X, y = synth1.get_data(N)
# Construct and fit the impact chart model:
impact_model = XGBoostImpactModel()
impact_model.fit(X, y)
# Plot the charts. The return value is a dictionary
# with one chart per column of X.
impact_charts = impact_model.impact_charts(X)
```
There are other options to control and format the charts, but the
code above is sufficient to get the job done.
For more details, please see the
`Synthetic Data.ipynb`
notebook, from which the code above is derived.
## How Do Impact Charts Work
Impact charts are built on top of
[shap](https://github.com/shap/shap),
which uses Shapley values to interpret predictions made by
machine learning models. For more details on how this is done,
please see
[this paper](https://datapinions.com/wp-content/uploads/2024/01/impactcharts.pdf)
and/or
[this blog post](https://datapinions.com/an-introduction-to-impact-charts/).
## More on Impact Charts
Applications built on top of `impactchart` be found in the
projects [evlcharts](https://github.com/vengroff/evlcharts) and
[rihcharts](https://github.com/vengroff/rihcharts).
An earlier version of the code that led to what is now
here produced the impact charts available at [http://rih.datapinions.com/impact.html](http://rih.datapinions.com/impact.html).
This work, and the motivation for the impact chart approach, is discussed at length in the blog post
[Using Interpretable Machine Learning to Analyze Racial and Ethnic Disparities in Home Values](https://datapinions.com/using-interpretable-machine-learning-to-analyze-racial-and-ethnic-disparities-in-home-values/).
Raw data
{
"_id": null,
"home_page": "https://github.com/vengroff/impactchart",
"name": "impactchart",
"maintainer": null,
"docs_url": null,
"requires_python": "<4.0,>=3.11",
"maintainer_email": null,
"keywords": "impact charts, regression, machine learning, analysis",
"author": "Darren Vengroff",
"author_email": null,
"download_url": "https://files.pythonhosted.org/packages/98/83/a5b1ae0eb026a3adc9ba9be0c81a0c51676455ef66a4dc96325194bef7c9/impactchart-0.6.0.tar.gz",
"platform": null,
"description": "# impactchart\n\n[](https://firstdonoharm.dev/version/3/0/cl-eco-extr-ffd-law-mil-sv.html)\n\n\n\n\n\n\n\n\n[](https://vengroff.github.io/impactchart/)\n[](https://vengroff.github.io/impactchart/)\n\n`impactchart` is a Python library for generating impact charts.\n\n## What are Impact Charts?\n\n(See also the [ACM FAccT paper](https://facctconference.org/static/papers24/facct24-80.pdf) for more background and details.)\n\nImpact charts make it easy to take a data set and visualize the impact of one variable \non another in ways that techniques like scatter plots and linear regression can't, \nespecially when there are other variables involved. \n\nFor example, consider this impact chart, which looks at the impact of \n`x_3` on `y` in a particular data set:\n\n\n\nThe green dots represent our best estimate of the \nimpact. The grey dots around them represent the estimate of the impact based on many\n(in this case 50) different machine learning models. When they are close to the green\ndots, as on the left side of the chart it means there is strong agreement among the\nmodels as to the impact. When they are farther apart, as on the left side of the chart,\nthere is less agreement.\n\nThe general shape of the curve of green dots, and the fact that the gray dots remain\nrather tightly grouped around it, suggest that the impact of `x_3` on `y` is very limited\nwhen `x_3` is negative. But as it becomes increasingly positive, it's impact grows more \nand more rapidly. It might even be exponential \\[Spoiler alert: this is synthetic data and \nthe impact is exponential.].\n\nNow let's compare what we see in the impact chart to a traditional scatter plot, which\nis a popular way of taking a first look at the relationship between variables.\nUsing the same data set, here is a scatter plot of `y` vs. `x_3`:\n\n\n\nIn the scatter plot, we also added regression curves for linear regression and\nquadratic regression, but they did not tell us much. The reason is that\nin the data set we are looking at, `x_3` isn't the only feature that impacts\n`y`. There are other `x_i` whose cumulative impact hides that of `x_3`.\n\nThe impact chart, on the other hand, we see the impact of `x_3` on `y` independent\nof the effect of any other `x_i`. \n\nThe reason impact charts like the one we are looking at are so powerful is that they \nvery clealy and directly show us the impact of one feature in a data set on a target \nof interest.\n\nImpact charts can find impact even though, unlike parametric regression techniques, \nthey don't have any _a priori_ knowledge of the shape of the impacts they are looking for. \nFor example, in the data set we have been looking at, there is another feature `x_2` whose \nimpact on `y` is sinusoidal. And the impact chart for it shows this clearly\n\n\n\nIn general, no matter what the shape of the impact of one variable on another, \nwhether it is polynomial, exponential, logarithmic, polylogarithmic, has seasonality \nor other repetitive structure, if the impact is there you are going\nto get a good visualization with an impact chart.\n\nAnd what if there is no impact? What if the two variables are not correlated at all.\nWe can see an example of that case in the following impact chart:\n\n\n\nIn this case, the impact chart is essentially flat at zero. There is a little bit\nof random noise resulting from the way we created and trained the underlying machine\nlearning models, but the fact that there `x_4` had negligible impact on `y`, at least\nrelative to the impacts `x_2` and `x_3` had, is quite clear.\n\n## What do Impact Charts on Real Data Look Like?\n\nThe data set we looked at above to introduce impact charts is synthetic. We deliberately\nconstructed it so some of the variables had known impacts and others had none. It was\npart of a test to show that the ideas behind impact charts worked and that the code \nthat we wrote to implement them was working.\n\nBut of course the real reason for impact charts is to look for impacts that we did not\ndeliberately put there for the code to find. So what do impact charts look like on real \ndata? One of the first data sets we tried it on consists of data about the median income of renters,\nthe racial and ethnic make up of the population of renters, and the rate of eviction at the \ncensus tract level in hundreds of communities around the United States. We made impact charts\nat the county level to show the impact of income, race, and ethnicity on eviction.\n\nHere is an example:\n\n\n\nThis chart looks at the impact of the median household income of renters (at the census tract level)\non the eviction rate in DeKalb County, Georgia in years between 2009 and 2018. The impact is measured\nin units of eviction cases filed per 100 renters. \n\nWe might hypothesize that the lowest income renters\nwould have the highest eviction rates. But this impact chart tells us otherwise. In very low income\ntracts, there is an impact around -5, meaning that these tracts have an eviction rate five points\nlower than they otherwise would. Why would that be? In some cases, those at the very bottom of the income\nscale live in public housing or voucher-supported housing (commonly known as Section-8). In these\nkind of settings, they have more protection against eviction than renters do on the open market.\n\nBetween $25,000 and $40,000, the impact is almost entirely above zero. These tracts experience higher\nrates of eviction because of their income level. This is consistent with a population of people\nwho have enough income to pay rent under normal circumstances but not enough to absorb the kind \nof shock (a major car repair or medical bill) that those with higher income might have the savings\nor credit to survive without being evicted. Above $50,000 in household income, the curve is mostly \nflat at zero impact. This doesn't mean none of these people ever get evicted. It just means that\nit is not their income that is driving the eviction. \n\nBut between $50,00 and $100,000, the green\ndots are bifircated into a zero impact group and a group around -3 to -4. Usually this means\nthere is some other variable that is causing the difference between the two, but it was not\npresent in our data set, so the models did the best they could in assigning that impact. If we could\nadd tha right variable to our data set, the effect would go away.\n\nIn just one chart, we can start to build a very interesting picture of the dynamics of the social and\neconomic processes behind the data. And we can compare the impact of variables. Without going into\nall of the details we discuss \n[elsewhere](https://datapinions.com/the-impact-of-demographics-and-income-on-eviction-rates/)\nhere is another impact chart from the same data set. But instead of looking at the impact of \nmedian renter income, we look at the impact of the fraction of renters in a census tract who\nidentify as Black alone. The chart is:\n\n\n\nCensus tracts where very few Black renters live have a negative impact, meaning their eviction\nrates are lower. The dats set include, income, whose impact we already looked at, so this\nimpact chart shows the impact that can be attributed to race independent of income, even if\nthe two are correlated to some degree. Note also that the two impact charts share the same\nvertical scale, so that the magnitudes of the impacts can be compared.\n\nIf you would like to see more impact charts generated from this data set, please visit\n[http://evl.datapinions.com/](http://evl.datapinions.com/).\n\n## How Do I Make an Impact Chart\n\nThe first step is to install the `impactchart` code in a virtual environment using\n\n```shell\npip install impactchart\n```\n\nFrom there, the simplest way to make your first impact chart is to \nreplicate the code that generated the first impact charts we saw above, using\nsynthetic data. It is as follows:\n\n```python\nfrom impactchart.model import XGBoostImpactModel\nfrom impactchartdemo.dataset import synth1\n\n# Generate the data set:\nN = 200\nX, y = synth1.get_data(N)\n\n# Construct and fit the impact chart model:\nimpact_model = XGBoostImpactModel()\nimpact_model.fit(X, y)\n\n# Plot the charts. The return value is a dictionary\n# with one chart per column of X.\nimpact_charts = impact_model.impact_charts(X)\n```\n\nThere are other options to control and format the charts, but the \ncode above is sufficient to get the job done.\n\nFor more details, please see the \n`Synthetic Data.ipynb` \nnotebook, from which the code above is derived.\n\n## How Do Impact Charts Work\n\nImpact charts are built on top of \n[shap](https://github.com/shap/shap), \nwhich uses Shapley values to interpret predictions made by\nmachine learning models. For more details on how this is done,\nplease see \n[this paper](https://datapinions.com/wp-content/uploads/2024/01/impactcharts.pdf) \nand/or \n[this blog post](https://datapinions.com/an-introduction-to-impact-charts/).\n\n## More on Impact Charts\n\nApplications built on top of `impactchart` be found in the \nprojects [evlcharts](https://github.com/vengroff/evlcharts) and\n[rihcharts](https://github.com/vengroff/rihcharts).\n\nAn earlier version of the code that led to what is now\nhere produced the impact charts available at [http://rih.datapinions.com/impact.html](http://rih.datapinions.com/impact.html).\nThis work, and the motivation for the impact chart approach, is discussed at length in the blog post\n[Using Interpretable Machine Learning to Analyze Racial and Ethnic Disparities in Home Values](https://datapinions.com/using-interpretable-machine-learning-to-analyze-racial-and-ethnic-disparities-in-home-values/).\n",
"bugtrack_url": null,
"license": "HL3-CL-ECO-EXTR-FFD-LAW-MIL-SV",
"summary": "A package for generating impact charts.",
"version": "0.6.0",
"project_urls": {
"Homepage": "https://github.com/vengroff/impactchart",
"Repository": "https://github.com/vengroff/impactchart"
},
"split_keywords": [
"impact charts",
" regression",
" machine learning",
" analysis"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "6c89cb5d786dd3e7bcc73c0d7f2085ec7cdf007c6e82deaa3108c6dab09aa202",
"md5": "2c2d8c9510662c0baa7643a6aade1983",
"sha256": "c2cca3266042fe4d3f0650f0c261a6e7c87bf9c771798e374a86153f8cca77ef"
},
"downloads": -1,
"filename": "impactchart-0.6.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "2c2d8c9510662c0baa7643a6aade1983",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": "<4.0,>=3.11",
"size": 30183,
"upload_time": "2024-07-16T01:23:28",
"upload_time_iso_8601": "2024-07-16T01:23:28.758123Z",
"url": "https://files.pythonhosted.org/packages/6c/89/cb5d786dd3e7bcc73c0d7f2085ec7cdf007c6e82deaa3108c6dab09aa202/impactchart-0.6.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "9883a5b1ae0eb026a3adc9ba9be0c81a0c51676455ef66a4dc96325194bef7c9",
"md5": "8f00c8712ddb0a7f39219a94d137ddfb",
"sha256": "3315fcbd68c805863df7e104e1ba55247f5cdcfccb98038002caac97affaba33"
},
"downloads": -1,
"filename": "impactchart-0.6.0.tar.gz",
"has_sig": false,
"md5_digest": "8f00c8712ddb0a7f39219a94d137ddfb",
"packagetype": "sdist",
"python_version": "source",
"requires_python": "<4.0,>=3.11",
"size": 31016,
"upload_time": "2024-07-16T01:23:30",
"upload_time_iso_8601": "2024-07-16T01:23:30.346033Z",
"url": "https://files.pythonhosted.org/packages/98/83/a5b1ae0eb026a3adc9ba9be0c81a0c51676455ef66a4dc96325194bef7c9/impactchart-0.6.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-07-16 01:23:30",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "vengroff",
"github_project": "impactchart",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "impactchart"
}
JSON
