Skip to content

Commit

Permalink
pymc marketing intro
Browse files Browse the repository at this point in the history
  • Loading branch information
@juanitorduz
juanitorduz committed Jan 11, 2025
1 parent f0a353e commit 61c5209
Show file tree
Hide file tree
Showing 8 changed files with 306 additions and 13 deletions.
189 changes: 181 additions & 8 deletions Presentations/amld_2025/amld_2025.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,71 @@
margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */
vertical-align: middle;
}
/* CSS for syntax highlighting */
pre > code.sourceCode { white-space: pre; position: relative; }
pre > code.sourceCode > span { line-height: 1.25; }
pre > code.sourceCode > span:empty { height: 1.2em; }
.sourceCode { overflow: visible; }
code.sourceCode > span { color: inherit; text-decoration: inherit; }
div.sourceCode { margin: 1em 0; }
pre.sourceCode { margin: 0; }
@media screen {
div.sourceCode { overflow: auto; }
}
@media print {
pre > code.sourceCode { white-space: pre-wrap; }
pre > code.sourceCode > span { display: inline-block; text-indent: -5em; padding-left: 5em; }
}
pre.numberSource code
{ counter-reset: source-line 0; }
pre.numberSource code > span
{ position: relative; left: -4em; counter-increment: source-line; }
pre.numberSource code > span > a:first-child::before
{ content: counter(source-line);
position: relative; left: -1em; text-align: right; vertical-align: baseline;
border: none; display: inline-block;
-webkit-touch-callout: none; -webkit-user-select: none;
-khtml-user-select: none; -moz-user-select: none;
-ms-user-select: none; user-select: none;
padding: 0 4px; width: 4em;
}
pre.numberSource { margin-left: 3em; padding-left: 4px; }
div.sourceCode
{ color: #e1e4e8; background-color: #24292e; }
@media screen {
pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
}
code span { color: #e1e4e8; } /* Normal */
code span.al { color: #ff5555; font-weight: bold; } /* Alert */
code span.an { color: #6a737d; } /* Annotation */
code span.at { color: #f97583; } /* Attribute */
code span.bn { color: #79b8ff; } /* BaseN */
code span.bu { color: #f97583; } /* BuiltIn */
code span.cf { color: #f97583; } /* ControlFlow */
code span.ch { color: #9ecbff; } /* Char */
code span.cn { color: #79b8ff; } /* Constant */
code span.co { color: #6a737d; } /* Comment */
code span.cv { color: #6a737d; } /* CommentVar */
code span.do { color: #6a737d; } /* Documentation */
code span.dt { color: #f97583; } /* DataType */
code span.dv { color: #79b8ff; } /* DecVal */
code span.er { color: #ff5555; text-decoration: underline; } /* Error */
code span.ex { color: #f97583; font-weight: bold; } /* Extension */
code span.fl { color: #79b8ff; } /* Float */
code span.fu { color: #b392f0; } /* Function */
code span.im { color: #9ecbff; } /* Import */
code span.in { color: #6a737d; } /* Information */
code span.kw { color: #f97583; } /* Keyword */
code span.op { color: #e1e4e8; } /* Operator */
code span.ot { color: #b392f0; } /* Other */
code span.pp { color: #f97583; } /* Preprocessor */
code span.re { color: #6a737d; } /* RegionMarker */
code span.sc { color: #79b8ff; } /* SpecialChar */
code span.ss { color: #9ecbff; } /* SpecialString */
code span.st { color: #9ecbff; } /* String */
code span.va { color: #ffab70; } /* Variable */
code span.vs { color: #9ecbff; } /* VerbatimString */
code span.wa { color: #ff5555; } /* Warning */
</style>
<link rel="stylesheet" href="amld_2025_files/libs/revealjs/dist/theme/quarto-4b2d1750cd98acf9ce31f523000b3309.css">
<link rel="stylesheet" href="amld_2025_files/style.css">
Expand Down Expand Up @@ -226,6 +291,10 @@ <h2>Saturation Effect</h2>
</div>

<img data-src="amld_2025_files/static/images/saturation.png" class="quarto-figure quarto-figure-center r-stretch" width="1000"></section>
<section id="media-transformations" class="slide level2">
<h2>Media Transformations</h2>

<img data-src="amld_2025_files/static/images/media_transformations.png" class="quarto-figure quarto-figure-center r-stretch" width="1000"></section>
<section id="additional-effects" class="slide level2">
<h2>Additional Effects</h2>

Expand All @@ -237,23 +306,127 @@ <h2>MMM as a Causal Model</h2>
<p><a href="https://www.pymc-marketing.io/en/latest/notebooks/mmm/mmm_roas.html">PyMC-Marketing Example: Unobserved Confounders, ROAS and Lift Tests</a></p>
</div>
</section>
<section id="why-bayesian-mmms" class="slide level2">
<section id="why-bayesian-mmms" class="slide level2 smaller">
<h2>Why Bayesian MMMs?</h2>
<h3 id="some-mmm-challenges">Some MMM Challenges</h3>
<div>
<ul>
<li class="fragment"><p>Limited data (typically 2-3 years of data, sometimes weekly granularity).</p></li>
<li class="fragment"><p>Media variables are generally very correlated.</p></li>
<li class="fragment"><p>Unobserved confounders (e.g.&nbsp;competitors investments).</p></li>
</ul>
</div>
<h3 id="bayesian-mmms">Bayesian MMMs</h3>
<div>
<ul>
<li>…</li>
<li class="fragment"><p>Uncertainty quantification.</p></li>
<li class="fragment"><p>Domain knowledge through priors.</p></li>
<li class="fragment"><p>Lift test calibration (e.g.&nbsp;geo-tests or switch-back experiments).</p></li>
<li class="fragment"><p>Time-varying parameters with Bayesian regularization (e.g.&nbsp;strong priors or hierarchies).</p></li>
<li class="fragment"><p>Risk-based budget optimization.</p></li>
</ul>
</div>
</section>
<section id="pymc-marketing" class="slide level2">
<h2>PyMC-Marketing</h2>
<ul>
<li>…</li>
</ul>

<img data-src="amld_2025_files/static/images/logos/marketing-logo-light.jpg" class="quarto-figure quarto-figure-center r-stretch" width="1000"><blockquote>
<p>Bayesian marketing toolbox in PyMC. Media Mix (MMM), customer lifetime value (CLV), buy-till-you-die (BTYD) models and more.</p>
</blockquote>
<div class="footer">
<p><a href="https://www.pymc-marketing.io/">PyMC-Marketing</a></p>
</div>
</section>
<section id="parameter-recovery-example" class="slide level2">
<h2>Parameter Recovery Example</h2>
<section id="pymc-marketing-api" class="slide level2">
<h2>PyMC-Marketing API</h2>
<div class="sourceCode" id="cb1" data-code-line-numbers="|1-3|4-11|13-23|25-36"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb1-1"><a href=""></a><span class="im">from</span> pymc_marketing.mmm <span class="im">import</span> MMM, GeometricAdstock, LogisticSaturation</span>
<span id="cb1-2"><a href=""></a><span class="im">from</span> pymc_marketing.prior <span class="im">import</span> Prior</span>
<span id="cb1-3"><a href=""></a></span>
<span id="cb1-4"><a href=""></a><span class="co"># Define priors</span></span>
<span id="cb1-5"><a href=""></a>my_model_config <span class="op">=</span> {</span>
<span id="cb1-6"><a href=""></a> <span class="st">"intercept"</span>: Prior(<span class="st">"Normal"</span>, mu<span class="op">=</span><span class="fl">0.5</span>, sigma<span class="op">=</span><span class="fl">0.1</span>),</span>
<span id="cb1-7"><a href=""></a> ...,</span>
<span id="cb1-8"><a href=""></a> <span class="st">"likelihood"</span>: Prior(</span>
<span id="cb1-9"><a href=""></a> <span class="st">"TruncatedNormal"</span>, lower<span class="op">=</span><span class="dv">0</span>, sigma<span class="op">=</span>Prior(<span class="st">"Exponential"</span>, lam<span class="op">=</span><span class="dv">1</span>)</span>
<span id="cb1-10"><a href=""></a> ),</span>
<span id="cb1-11"><a href=""></a>}</span>
<span id="cb1-12"><a href=""></a></span>
<span id="cb1-13"><a href=""></a><span class="co"># Define the model</span></span>
<span id="cb1-14"><a href=""></a>mmm <span class="op">=</span> MMM(</span>
<span id="cb1-15"><a href=""></a> model_config<span class="op">=</span>my_model_config,</span>
<span id="cb1-16"><a href=""></a> date_column<span class="op">=</span><span class="st">"date_week"</span>,</span>
<span id="cb1-17"><a href=""></a> adstock<span class="op">=</span>GeometricAdstock(l_max<span class="op">=</span><span class="dv">8</span>),</span>
<span id="cb1-18"><a href=""></a> saturation<span class="op">=</span>LogisticSaturation(),</span>
<span id="cb1-19"><a href=""></a> channel_columns<span class="op">=</span>channel_columns,</span>
<span id="cb1-20"><a href=""></a> control_columns<span class="op">=</span>control_columns,</span>
<span id="cb1-21"><a href=""></a> time_varying_intercept<span class="op">=</span><span class="va">True</span>,</span>
<span id="cb1-22"><a href=""></a> yearly_seasonality<span class="op">=</span><span class="dv">2</span>,</span>
<span id="cb1-23"><a href=""></a>)</span>
<span id="cb1-24"><a href=""></a></span>
<span id="cb1-25"><a href=""></a><span class="co"># Fit the model</span></span>
<span id="cb1-26"><a href=""></a>_ <span class="op">=</span> mmm.fit(</span>
<span id="cb1-27"><a href=""></a> X<span class="op">=</span>X,</span>
<span id="cb1-28"><a href=""></a> y<span class="op">=</span>y,</span>
<span id="cb1-29"><a href=""></a> target_accept<span class="op">=</span><span class="fl">0.85</span>,</span>
<span id="cb1-30"><a href=""></a> chains<span class="op">=</span><span class="dv">4</span>,</span>
<span id="cb1-31"><a href=""></a> draws<span class="op">=</span><span class="dv">2_000</span>,</span>
<span id="cb1-32"><a href=""></a> nuts_sampler<span class="op">=</span><span class="st">"numpyro"</span>, <span class="co"># &lt;- We can use many samplers and backends!</span></span>
<span id="cb1-33"><a href=""></a> random_seed<span class="op">=</span>rng,</span>
<span id="cb1-34"><a href=""></a>)</span>
<span id="cb1-35"><a href=""></a><span class="co"># Sample from the posterior predictive distribution</span></span>
<span id="cb1-36"><a href=""></a>_ <span class="op">=</span> mmm.sample_posterior_predictive(X, extend_idata<span class="op">=</span><span class="va">True</span>, combined<span class="op">=</span><span class="va">True</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
</section>
<section id="prior-specification" class="slide level2">
<h2>Prior Specification</h2>

<img data-src="amld_2025_files/static/images/beta_priors.png" class="quarto-figure quarto-figure-center r-stretch" width="1000"><div class="footer">
<p><a href="https://www.pymc-marketing.io/en/latest/notebooks/general/prior_predictive.html">PyMC-Marketing Example: Prior Predictive Modeling</a></p>
</div>
</section>
<section id="attribution-decomposition" class="slide level2">
<h2>Attribution Decomposition</h2>
<div class="sourceCode" id="cb2"><pre class="sourceCode numberSource python number-lines code-with-copy"><code class="sourceCode python"><span id="cb2-1"><a href=""></a>mmm.plot_components_contributions(original_scale<span class="op">=</span><span class="va">True</span>)<span class="op">;</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>

<img data-src="amld_2025_files/static/images/model_components.png" class="quarto-figure quarto-figure-center r-stretch" width="1000"></section>
<section id="return-on-ad-spend-roas" class="slide level2">
<h2>Return on Ad Spend (ROAS)</h2>

<img data-src="amld_2025_files/static/images/roas_posterior.png" class="quarto-figure quarto-figure-center r-stretch" width="1000"></section>
<section id="out-of-sample-forecasting" class="slide level2 smaller">
<h2>Out-of-sample Forecasting</h2>
<div class="columns">
<div class="column" style="width:50%;">
<div class="quarto-figure quarto-figure-center">
<figure>
<p><img data-src="amld_2025_files/static/images/tscv.png" class="quarto-figure quarto-figure-center" width="500"></p>
</figure>
</div>
</div><div class="column" style="width:50%;">
<div title="Model Evaluation Framework">
<div class="callout callout-tip callout-titled callout-style-default">
<div class="callout-body">
<div class="callout-title">
<div class="callout-icon-container">
<i class="callout-icon"></i>
</div>
<p><strong>Model Evaluation Framework</strong></p>
</div>
<div class="callout-content">
<p>Evaluate the model on the holdout set:</p>
<ul>
<li>…</li>
<li>Prediction accuracy (CRPS)</li>
<li>Parameter Stability</li>
</ul>
</div>
</div>
</div>
</div>
<div class="quarto-figure quarto-figure-center">
<figure>
<p><img data-src="amld_2025_files/static/images/tscv_alpha.png" class="quarto-figure quarto-figure-center" width="900"></p>
</figure>
</div>
</div></div>

</section>
</div>
Expand Down
130 changes: 125 additions & 5 deletions Presentations/amld_2025/amld_2025.qmd
View file
Original file line number Diff line number Diff line change
Expand Up @@ -90,6 +90,9 @@ $$

![](amld_2025_files/static/images/saturation.png){fig-align="center" width="1000"}

## Media Transformations

![](amld_2025_files/static/images/media_transformations.png){fig-align="center" width="1000"}

## Additional Effects

Expand All @@ -103,15 +106,132 @@ $$
[PyMC-Marketing Example: Unobserved Confounders, ROAS and Lift Tests](https://www.pymc-marketing.io/en/latest/notebooks/mmm/mmm_roas.html)
:::

## Why Bayesian MMMs?
## Why Bayesian MMMs? {.smaller}

### Some MMM Challenges

::: incremental

- Limited data (typically 2-3 years of data, sometimes weekly granularity).

- Media variables are generally very correlated.

- Unobserved confounders (e.g. competitors investments).

:::

### Bayesian MMMs

::: incremental

- Uncertainty quantification.

- ...
- Domain knowledge through priors.

- Lift test calibration (e.g. geo-tests or switch-back experiments).

- Time-varying parameters with Bayesian regularization (e.g. strong priors or hierarchies).

- Risk-based budget optimization.

:::

## PyMC-Marketing

- ...
![](amld_2025_files/static/images/logos/marketing-logo-light.jpg){fig-align="center" width="1000"}

> Bayesian marketing toolbox in PyMC. Media Mix (MMM), customer lifetime value (CLV), buy-till-you-die (BTYD) models and more.
::: footer
[PyMC-Marketing](https://www.pymc-marketing.io/)
:::

## PyMC-Marketing API

```{.python code-line-numbers="|1-3|4-11|13-23|25-36"}
from pymc_marketing.mmm import MMM, GeometricAdstock, LogisticSaturation
from pymc_marketing.prior import Prior
# Define priors
my_model_config = {
"intercept": Prior("Normal", mu=0.5, sigma=0.1),
...,
"likelihood": Prior(
"TruncatedNormal", lower=0, sigma=Prior("Exponential", lam=1)
),
}
# Define the model
mmm = MMM(
model_config=my_model_config,
date_column="date_week",
adstock=GeometricAdstock(l_max=8),
saturation=LogisticSaturation(),
channel_columns=channel_columns,
control_columns=control_columns,
time_varying_intercept=True,
yearly_seasonality=2,
)
# Fit the model
_ = mmm.fit(
X=X,
y=y,
target_accept=0.85,
chains=4,
draws=2_000,
nuts_sampler="numpyro", # <- We can use many samplers and backends!
random_seed=rng,
)
# Sample from the posterior predictive distribution
_ = mmm.sample_posterior_predictive(X, extend_idata=True, combined=True)
```

## Prior Specification

![](amld_2025_files/static/images/beta_priors.png){fig-align="center" width="1000"}

::: footer
[PyMC-Marketing Example: Prior Predictive Modeling](https://www.pymc-marketing.io/en/latest/notebooks/general/prior_predictive.html)
:::

## Attribution Decomposition

```{.python}
mmm.plot_components_contributions(original_scale=True);
```

![](amld_2025_files/static/images/model_components.png){fig-align="center" width="1000"}


## Return on Ad Spend (ROAS)

![](amld_2025_files/static/images/roas_posterior.png){fig-align="center" width="1000"}

## Out-of-sample Forecasting {.smaller}

::: {.columns}

::: {.column width="50%"}

## Parameter Recovery Example
![](amld_2025_files/static/images/tscv.png){fig-align="center" width="500"}

- ...
:::


::: {.column width="50%"}

::: {.callout-tip title="Model Evaluation Framework"}

Evaluate the model on the holdout set:

- Prediction accuracy (CRPS)
- Parameter Stability

:::

![](amld_2025_files/static/images/tscv_alpha.png){fig-align="center" width="900"}

:::

:::
Binary file added Presentations/amld_2025/amld_2025_files/static/images/beta_priors.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added Presentations/amld_2025/amld_2025_files/static/images/media_transformations.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added Presentations/amld_2025/amld_2025_files/static/images/model_components.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added Presentations/amld_2025/amld_2025_files/static/images/roas_posterior.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added Presentations/amld_2025/amld_2025_files/static/images/tscv.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added Presentations/amld_2025/amld_2025_files/static/images/tscv_alpha.png
View file
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.

0 comments on commit 61c5209

Please sign in to comment.