Dynamic-kernels

This repository contains the code for the paper:
"Context-aware machine learning with dynamically assembled weight matrices."
preprint: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4182104

Code

All the code is written in the newest stable tensorflow version v2.5.0.
The code is designed with to be compatible with the keras functional, sequantial & model API.

Project tree

• Src
  • Optimizers
  • Layers
  • Models
  • Callbacks
  • Initializers
• Notebooks

Example

The following is a sample for code usage.

from src.layers import CAAddDense
from src.optimizers import SVDAdam
from src.models.utils import wrap_model

# Create a dataset
data = tf.data.Dataset.from_generator(...)
# Make a context model
context_inputs = tf.keras.layers.Inputs(...)
context_outputs = tf.keras.layers.Dense(...)(context_inputs)
context_model = tf.keras.models.Model()(outputs=context_outputs, inputs=context_inputs)
# Create a model
inputs = tf.keras.layers.Inputs(...)
context = tf.keras.layers.Inputs(...)
context = context_model(context)
hidden = SVDDense(...)([inputs, context])
... # Add more complicated architecture
outputs = tf.keras.layers.Dense(...)(hidden)
# Make model
model = tf.keras.Models(outputs=outputs, inputs=(inputs, context))
# Wrap model such that dataset is given to optimizer
model = wrap_model(model)
# Create optimizer --> Needs a model and optional context model to be created
optimizer = SVDAdam(model, context_model, ...)
# Create loss object
loss_fn = tf.keras.losses.MeanSquaredError()
# Compile model
Model.compile(optimizer, loss_fn, ...)
# train model
model.fit(data, ...)

More detailed examples, corresponding to the experiment section in the paper, can be found in the notebooks.

The following notebooks are available:

• Artificial2D contains a two dimensional example for the artificial data.
• ArtificialMultiDim contains higher dimensional modeling and analysis for the artificial data.
• Indian Covid Data - Baseline contains time series analysis and modeling for the per state individual models.
• Indian Covid Data contains time series analysis and modeling corresponding to the general models.
• DataAnalysis contains the code used for scraping of the image tags for the YFCCM100-GEO100 dataset.
• GeoImagesModel contains models and analysis used for the the YFCCM100-GEO100 dataset.

Data

Data corresponding to the indian covid time series: Coming soon
Data corresponding to the YFCCM100-GEO100 and tags data: Coming soon