Module for transforming a Meridian model into a structured MMM schema.
This module provides the ModelKernelProcessor
, which is responsible for
transforming the internal state of a trained Meridian model object into a
structured and portable format defined by the MmmKernel
protobuf message.
The "kernel" includes essential information about the model, such as:
- Model specifications and hyperparameters.
- Inferred parameters distributions (as a serialized ArViz inference data).
- MMM-agnostic marketing data (i.e. input data to the model).
This serialized representation allows the model to be saved, loaded, and
analyzed across different environments or by other tools that understand the MmmKernel
schema.
The serialization logic is primarily handled by the MeridianSerde
class from
the schema.serde
package.
Key Classes:
-
ModelKernelProcessor: The processor class that takes a Meridian model instance and populates anMmmKernelmessage.
Example Usage:
import
meridian
from
meridian.model
import
model
from
mmm.v1
import
mmm_pb2
from
schema.processors
import
model_kernel_processor
import
semver
# Assuming 'mmm' is a `meridian.model.Meridian` object.
# Example:
# mmm = meridian.model.Meridian(...)
# mmm.sample_prior(...)
# mmm.sample_posterior(...)
processor
=
model_kernel_processor
.
ModelKernelProcessor
(
meridian_model
=
mmm
,
model_id
=
"my_model_v1"
,
)
# Create an output Mmm proto message
output_proto
=
mmm_pb2
.
Mmm
()
# Populate the mmm_kernel field
processor
(
output_proto
)
# Now output_proto.mmm_kernel contains the serialized model.
# This can be saved to a file, sent over a network, etc.
print
(
f
"Model Kernel ID:
{
output_proto
.
mmm_kernel
.
model_id
}
"
)
print
(
f
"Meridian Version:
{
output_proto
.
mmm_kernel
.
meridian_version
}
"
)
# Access other fields within output_proto.mmm_kernel as needed.
Classes
class ModelKernelProcessor
: Transcribes a model's stats into an "MmmKernel
message.
