Opsmo.CRPM (opsmo v0.3.4)
Logistic Regression model to Compute Resource Placement in a Cluster.
Input
It takes an input with the following
- Requested CPU
- Requested Memory
- Requested Disk
- Available CPU
- Available Memory
- Available Disk
[requested_cpu, requested_memory, requested_disk, available_cpu, available_memory, available_disk]
Each input value should be the normalized value between 0 and 1.
[0.05, 0.0625, 0.004, 0.825, 0.65, 0.55]
[0.05, 0.0625, 0.004, 0.314, 0.55, 0.45]
[0.05, 0.0625, 0.004, 0.954, 0.35, 0.55]The above represents 3 nodes in the cluster. You can also add more nodes to the input if you have them. The requested resource is the same for the 3 nodes. What this does is it will ask the model to tell you the best node to place the requsted resource.
Usage
This model computes the placement of resources in a cluster. It takes a 6-dimensional input and outputs a 2-dimensional vector.
Summary
Functions
Instantiate a new model.
Predicts the placement score for given input resources.
Train the model.
Functions
See Opsmo.dump/2.
See Opsmo.load/1.
Instantiate a new model.
Predicts the placement score for given input resources.
Parameters
- model: The Axon model to use for prediction
- state: The trained model state containing weights and biases
- input: A list of 6 numbers representing [requested_cpu, requested_memory, requested_disk, available_cpu, available_memory, available_disk]
Returns
A 2-element list containing the prediction scores.
Example
iex> model = CRPM.model()
iex> state = CRPM.load()
iex> CRPM.predict(model, state, [0.05, 0.0625, 0.004, 0.825, 0.65, 0.55])
[0.8234, 0.1766]Train the model.
You can pass it a data stream or enumerable of data.
How to use
alias Opsmo.CRPM
data = CRPM.Dataset.train()
CRPM.train(data)