Example Model ParamsΒΆ
These are used in our Hot Gym examples. This raw model params file is used in the Quick Start.
# Type of model that the rest of these parameters apply to.
model: HTMPrediction
# Version that specifies the format of the config.
version: 1
# The section "aggregationInfo" specifies what field to aggregate with which
# aggregation method.
#
# Example of how to aggregate the field "consumption" with the method "mean"
# and the field "gym" with the method "first". Both field will be
# aggregregated over a period of 1h 15m, according to their respective
# aggregation methods.
#
# aggregationInfo:
# fields:
# - [consumption, sum]
# - [gym, first]
# minutes: 15
# hours: 1
#
# See nupic.data.aggregator for more info about supported aggregation methods.
aggregationInfo:
# "fields" should be a list of pairs. Each pair is a field name and an
# aggregation function (e.g. sum). The function will be used to aggregate
# multiple values of this field over the aggregation period.
fields:
- [consumption, mean]
# If a time unit is not listed, 0 will be its default value.
microseconds: 0
milliseconds: 0
minutes: 0
months: 0
seconds: 0
hours: 1
days: 0
weeks: 0
years: 0
predictAheadTime: null
# Parameters of the model to be created.
modelParams:
# The type of inference that this model will perform.
# Supported values are :
# - TemporalNextStep
# - TemporalClassification
# - NontemporalClassification
# - TemporalAnomaly
# - NontemporalAnomaly
# - TemporalMultiStep
# - NontemporalMultiStep
inferenceType: TemporalMultiStep
# Parameters of the Sensor region
sensorParams:
# Sensor diagnostic output verbosity control:
# - verbosity == 0: silent
# - verbosity in [1 .. 6]: increasing level of verbosity
verbosity: 0
# List of encoders and their parameters.
encoders:
consumption:
fieldname: consumption
name: consumption
resolution: 0.88
seed: 1
type: RandomDistributedScalarEncoder
timestamp_timeOfDay:
fieldname: timestamp
name: timestamp_timeOfDay
timeOfDay: [21, 1]
type: DateEncoder
timestamp_weekend:
fieldname: timestamp
name: timestamp_weekend
type: DateEncoder
weekend: 21
# The "sensorAutoReset" specifies the period for automatically generated
# resets from a RecordSensor.
#
# If None, disable automatically generated resets. Also disable for all
# values that evaluate to 0. Example:
# sensorAutoReset: null
#
#
# Valid keys for the "sensorAutoReset" option:
# sensorAutoReset:
# days: <int>
# hours: <int>
# minutes: <int>
# seconds: <int>
# milliseconds: <int>
# microseconds: <int>
# weeks: <int>
#
# Example for an automated reset every 1.5 days:
# sensorAutoReset:
# days: 1
# hours: 12
#
sensorAutoReset: null
# Controls whether the Spatial Pooler (SP) region is enabled.
spEnable: true
# Parameters of the SP region.
spParams:
# SP diagnostic output verbosity control:
# - spVerbosity == 0: silent
# - spVerbosity in [1..6]: increasing levels of verbosity
spVerbosity: 0
# Spatial Pooler implementation selector.
# Values can be: "py" or "cpp" (speed optimized)
spatialImp: cpp
# If 1, then during inhibition phase the winning columns are selected
# as the most active columns from the region as a whole. Otherwise, the
# winning columns are selected with respect to their local neighborhoods.
# Using global inhibition boosts performance x60.
globalInhibition: 1
# The desired density of active columns within a local inhibition area
# (the size of which is set by the internally calculated inhibitionRadius,
# which is in turn determined from the average size of the connected
# potential pools of all columns). The inhibition logic will insure that
# at most N columns remain ON within a local inhibition area, where
# N = localAreaDensity * (total number of columns in inhibition area).
localAreaDensity: -1.0
# Number of mini-columns in the region. This values needs to be the same
# as "columnCount" for the TM region (see "columnCount" under "tmParams").
columnCount: 2048
# Input size of the SP region. Must be the same as the output size of the
# previous region, that feeds into the SP region.
inputWidth: 946
# numActiveColumnsPerInhArea controls column inhibition in the SP region.
# It sets the maximum number of active columns in the SP region's output.
# If more mini-columns than "numActiveColumnsPerInhArea" are active then
# the weaker columns are suppressed.
# In the case where globalInhibition is turned on (only one inhibition area)
# numActiveColumnsPerInhArea should be set to 40. Because 40 means roughly
# 2% active mini-columns. Given that there is only one inhibition area:
# 40 / 2048 = 0.0195 so ~2% active mini-columns.
numActiveColumnsPerInhArea: 40
# An alternate way to control the density of the active columns. If
# numActiveColumnsPerInhArea is specified then localAreaDensity must be
# less than 0, and vice versa. When using numActiveColumnsPerInhArea, the
# inhibition logic will insure that at most 'numActiveColumnsPerInhArea'
# columns remain ON within a local inhibition area (the size of which is
# set by the internally calculated inhibitionRadius, which is in turn
# determined from the average size of the connected receptive fields of all
# columns). When using this method, as columns learn and grow their
# effective receptive fields, the inhibitionRadius will grow, and hence the
# net density of the active columns will *decrease*. This is in contrast to
# the localAreaDensity method, which keeps the density of active columns
# the same regardless of the size of their receptive fields.
numActiveColumnsPerInhArea: 40
# Random number generator seed.
# The seed affects the random aspects of initialization like the initial
# permanence values and which input bits are in the potential pool for
# each column. A fixed value ensures a reproducible result.
seed: 1956
# "potentialPct" controls what percentage of the columns's receptive field
# is available for potential synapses.
potentialPct: 0.85
# The default threshold for synapses. Any synapse whose permanence
# value is above the connected threshold is a "connected synapse", meaning
# it can contribute to the cell's firing. Typical value is 0.10.
synPermConnected: 0.1
# The amount by which an active synapse is incremented in each round.
# For active columns, the permanences to all active input bits in the
# columns potential pool are incremented by this amount.
synPermActiveInc: 0.04
# How quickly synapses grow and degrade.
# For active columns, the permanences to all inactive input bits in the
# columns potential pool are decremented by this amount.
synPermInactiveDec: 0.005
# "boostStrength" controls the strength of boosting. It should be a
# a float greater or equal than 0.0. No boosting is applied if
# boostStrength == 0.0. Boosting encourages efficient usage of SP columns.
boostStrength: 3.0
# Controls whether the Temporal Memory (TM) region is enabled.
tmEnable: true
# Parameters of the TM region.
tmParams:
# Controls the verbosity of the TM diagnostic output:
# - verbosity == 0: silent
# - verbosity in [1..6]: increasing levels of verbosity
# See verbosity in nupic.algorithms.temporaly_memory for more details.
verbosity: 0
# Number of mini-columns in the region. This values needs to be the same
# as "columnCount" for the SP region (see "columnCount" under "spParams").
columnCount: 2048
# The number of cells per column.
cellsPerColumn: 32
# Size of the TM region input. Must match the output width of the
# previous region.
inputWidth: 2048
# Random number generator seed.
# The seed affects the random aspects of initialization like the initial
# permanence values and which input bits are in the potential pool for
# each column. A fixed value ensures a reproducible result.
seed: 1960
# Temporal implementation of the TM region.
# Valid values are: 'py', 'cpp', 'tm_py', 'monitored_tm_py'
temporalImp: cpp
# The max number of synapses added to a segment during learning
newSynapseCount: 20
# Initial permanence for newly created synapses
initialPerm: 0.21
# Active synapses get their permanence counts incremented by permanenceInc.
# All other synapses get their permanence counts decremented
# by permanenceDec.
permanenceInc: 0.1
permanenceDec: 0.1
# maxAge controls globalDecay. Global decay will only decay segments
# that have not been activated for maxAge iterations, and will
# only do the global decay loop every maxAge iterations. The
# default (maxAge=1) reverts to the behavior where global decay
# is applied every iteration to every segment. Using maxAge > 1
# can significantly speed up the TM when global decay is used.
maxAge: 0
globalDecay: 0.0
# Note that maxSegmentsPerCell and maxSynapsesPerSegment cannot be set at
# the same time as globalDecay and maxAge.
maxSynapsesPerSegment: 32
maxSegmentsPerCell: 128
# Minimum number of active synapses for a segment to be considered
# during search for the best-matching segments.
# None=use default
# Replaces: tpMinThreshold
minThreshold: 12
# Segment activation threshold.
# A segment is active if it has >= activationThreshold connected synapses
# that are active. If set to null, then use default value.
activationThreshold: 16
# Can be one of the following: 'normal', 'activeState',
# 'activeState1CellPerCol'.
# - 'normal': output the OR of the active and predicted state.
# - 'activeState': output only the active state.
# - 'activeState1CellPerCol': output only the active state, and at
# most 1 cell/column. If more than 1 cell is
# active in a column, the one with the highest
# confidence is sent up.
# Default value is 'normal'.
outputType: normal
# "Pay Attention Mode" length. This tells the TM Region how many new
# elements to append to the end of a learned sequence at a time.
# Smaller values are better for datasets with short sequences,
# higher values are better for datasets with long sequences.
pamLength: 1
clParams:
# Classifier diagnostic output verbosity control;
# - verbosity == 0: silent
# - verbosity in [1..6]: increasing levels of verbosity
verbosity: 0
# Name of the Classifier region.
regionName: SDRClassifierRegion
# This controls how fast the classifier learns/forgets. Higher
# values make it adapt faster and forget older patterns faster.
alpha: 0.1
# This is set after the call to updateConfigFromSubConfig and is
# computed from the aggregationInfo and predictAheadTime.
steps: '1,5'
# maxCategoryCount reprensents the max number of categories / bucket indices
# of the field being learned on.
# In this case, we are not learning a category field, but we are trying to
# predict the next value of a continous field. So maxCategoryCount is the
# max number of buckets used by the encoder of this field. We set
# maxCategoryCount to 1000, an arbitrarily large number, to accomodate the
# max number of buckets created by the encoder.
# In a case where the classifier is learning a category field with
# 3 categories, then maxCategoryCount should be set to 3 (higher values are
# also OK).
maxCategoryCount: 1000
# Values can be: "py" or "cpp" (speed optimized)
implementation: cpp
# If set, don't create the SP network unless the user requests SP metrics.
trainSPNetOnlyIfRequested: false
