Models YAML Specification

Spice ML Models YAML manifest Specification v0.2 (beta)

Spice ML Models are defined in YAML. To create a Model, add a manifest YAML file to the /models path of a connected GitHub repository.

YAML Schema Reference

name

The name of the model. Only alpha characters and underscores are supported with no spaces.

description [optional]

An optional description for the model.

family [optional]

The model family.

type

The model type. For more details see Model Type Specification.

metadata

A key-value bag of metadata/parameters for the model. When type is autogluon or darts, model, is a reserved key to specify hyperparameters of the underlying model (see Model Type Specification)

handler

The code file that contains the entry points for training and inferencing.

E.g. gas_fees.py

epochs

Training iterations count.

training

Optional. Defines training for the model.

Example

training:
  entry_point: gas_fees.train
  sql: 'WITH counts AS (    SELECT block_number, count(1) as "count" FROM eth.recent_transactions GROUP BY block_number  )  SELECT number as "ts", CAST(b.base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(c."count" AS DOUBLE) as "y2"  FROM eth.recent_blocks b  INNER JOIN counts c ON b.number = c.block_number  WHERE b.base_fee_per_gas IS NOT NULL ORDER BY block_number DESC LIMIT 500'
  loggers:
    - tensorboard

entry_point

The entry point (function) for training.

E.g. gas_fees.train

sql

The Spice.ai SQL query for fetching training data.

E.g. WITH counts AS ( SELECT block_number, count(1) as "count" FROM eth.recent_transactions GROUP BY block_number ) SELECT number as "ts", CAST(b.base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(c."count" AS DOUBLE) as "y2" FROM eth.recent_blocks b INNER JOIN counts c ON b.number = c.block_number WHERE b.base_fee_per_gas IS NOT NULL ORDER BY block_number DESC LIMIT 500

loggers

A list of loggers to enable for each training run.

Supported loggers and their values are:

Logger	Value
TensorBoard	tensorboard

Example

training:
  loggers:
    - tensorboard

inferencing

Optional. Defines inferencing for the model.

Example

inferencing:
  entry_point: gas_fees.train
  sql: 'WITH counts AS (    SELECT block_number, count(1) as "count" FROM eth.recent_transactions GROUP BY block_number  )  SELECT number as "ts", CAST(b.base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(c."count" AS DOUBLE) as "y2"  FROM eth.recent_blocks b  INNER JOIN counts c ON b.number = c.block_number  WHERE b.base_fee_per_gas IS NOT NULL ORDER BY block_number DESC LIMIT 500'
  lookback_size: 30
  forecast_size: 1

entry_point

The entry point (function) for inferencing.

E.g. gas_fees.infer

sql

The Spice.ai SQL query for fetching inferencing data.

E.g. SELECT number as "ts", CAST(base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(transaction_count AS DOUBLE) as "y2" from eth.recent_blocks WHERE base_fee_per_gas IS NOT NULL ORDER BY ts DESC LIMIT 35

lookback_size

The lookback size/window.

forecast_size

The forecast size/window.

Example definition

# /models/gas_fees/model.yml
name: v0.1
family: gas_fees
description: A Ethereum gas fee forecasting model.
type: gasfees_v1
handler: gas_fees.py
epochs: 1
training:
  entry_point: gas_fees.train
  sql: 'WITH counts AS (    SELECT block_number, count(1) as "count" FROM eth.recent_transactions GROUP BY block_number  )  SELECT number as "ts", CAST(b.base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(c."count" AS DOUBLE) as "y2"  FROM eth.recent_blocks b  INNER JOIN counts c ON b.number = c.block_number  WHERE b.base_fee_per_gas IS NOT NULL ORDER BY block_number DESC LIMIT 500'
inferencing:
  entry_point: gas_fees.infer
  sql: 'SELECT number as "ts", CAST(base_fee_per_gas / 1000000000.0 AS DOUBLE) as "y", CAST(transaction_count AS DOUBLE) as "y2" from eth.recent_blocks WHERE base_fee_per_gas IS NOT NULL ORDER BY ts DESC LIMIT 35'
  lookback_size: 30
  forecast_size: 1
metadata:
  firecache: true
  covariate: true