Hmm, those generally look pretty good to me. You could try these two: 1) Change cellsPerColumn down to 16. This is probably the safest change. You could even maybe try 8.
2) The other set to try is lower the number of columns but still keep 2% sparsity: Set columnCount to 1024 (need to change for SP and TP) Also set numActiveColumnsPerInhArea to 20 I am not sure how well this will work. If it runs a lot faster, then you might then need to spend some time to tune some of the other parameters in the TP, such as newSynapseCount. Let us know if this worked! —Subutai On Tue, Oct 21, 2014 at 2:15 PM, Nicolas Thiébaud <[email protected]> wrote: > Thanks Subutai, this is I have been using: > > MODEL_PARAMS = { > # Type of model that the rest of these parameters apply to. > 'model': "CLA", > > # Version that specifies the format of the config. > 'version': 1, > > # Intermediate variables used to compute fields in modelParams and also > # referenced from the control section. > #'aggregationInfo': { 'days': 0, > # 'fields': [(u'c1', 'sum'), (u'c0', 'first')], > # 'hours': 1, > # 'microseconds': 0, > # 'milliseconds': 0, > # 'minutes': 0, > # 'months': 0, > # 'seconds': 0, > # 'weeks': 0, > # 'years': 0}, > > 'predictAheadTime': None, > > # Model parameter dictionary. > 'modelParams': { > # The type of inference that this model will perform > 'inferenceType': 'TemporalAnomaly', > > 'sensorParams': { > # Sensor diagnostic output verbosity control; > # if > 0: sensor region will print out on screen what it's > sensing > # at each step 0: silent; >=1: some info; >=2: more info; > # >=3: even more info (see compute() in > py/regions/RecordSensor.py) > 'verbosity' : 0, > > # Example: > # dsEncoderSchema = [ > # DeferredDictLookup('__field_name_encoder'), > # ], > # > # (value generated from DS_ENCODER_SCHEMA) > 'encoders': { > u'timestamp_timeOfDay': None, > u'timestamp_dayOfWeek': None, > u'timestamp_weekend': None, > u'channel1': { > 'clipInput': True, > 'fieldname': u'channel1', > 'maxval': 4.0, > 'minval': -4.0, > 'n': 134, > 'name': u'channel1', > 'type': 'ScalarEncoder', > 'w': 21 > }, > }, > > # A dictionary specifying the period for > automatically-generated > # resets from a RecordSensor; > # > # None = disable automatically-generated resets (also disabled > if > # all of the specified values evaluate to 0). > # Valid keys is the desired combination of the following: > # days, hours, minutes, seconds, milliseconds, microseconds, > weeks > # > # Example for 1.5 days: sensorAutoReset = > dict(days=1,hours=12), > 'sensorAutoReset' : None, > }, > > 'spEnable': True, > > 'spParams': { > # SP diagnostic output verbosity control; > # 0: silent; >=1: some info; >=2: more info; > 'spVerbosity' : 0, > > # Spatial Pooler implementation selector. > # Options: 'py', 'cpp' (speed optimized, new) > 'spatialImp' : 'cpp', > > 'globalInhibition': 1, > > # Number of columns in the SP (must be same as in TP) > 'columnCount': 2048, > > 'inputWidth': 0, > > # SP inhibition control (absolute value); > # Maximum number of active columns in the SP region's output > (when > # there are more, the weaker ones are suppressed) > 'numActiveColumnsPerInhArea': 40, > > 'seed': 1956, > > # potentialPct > # What percent of the columns's receptive field is available > # for potential synapses. > 'potentialPct': 0.8, > > # The default connected threshold. 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, > > 'synPermActiveInc': 0.0001, > > 'synPermInactiveDec': 0.0005, > > 'maxBoost': 1.0, > }, > > # Controls whether TP is enabled or disabled; > # TP is necessary for making temporal predictions, such as > predicting > # the next inputs. Without TP, the model is only capable of > # reconstructing missing sensor inputs (via SP). > 'tpEnable' : True, > > 'tpParams': { > # TP diagnostic output verbosity control; > # 0: silent; [1..6]: increasing levels of verbosity > # (see verbosity in nta/trunk/py/nupic/research/TP.py and > TP10X*.py) > 'verbosity': 0, > > # Number of cell columns in the cortical region (same number > for > # SP and TP) > # (see also tpNCellsPerCol) > 'columnCount': 2048, > > # The number of cells (i.e., states), allocated per column. > 'cellsPerColumn': 32, > > 'inputWidth': 2048, > > 'seed': 1960, > > # Temporal Pooler implementation selector (see _getTPClass in > # CLARegion.py). > 'temporalImp': 'cpp', > > # New Synapse formation count > # NOTE: If None, use spNumActivePerInhArea > # > # TODO: need better explanation > 'newSynapseCount': 20, > > # Maximum number of synapses per segment > # > 0 for fixed-size CLA > # -1 for non-fixed-size CLA > # > # TODO: for Ron: once the appropriate value is placed in TP > # constructor, see if we should eliminate this parameter from > # description.py. > 'maxSynapsesPerSegment': 32, > > # Maximum number of segments per cell > # > 0 for fixed-size CLA > # -1 for non-fixed-size CLA > # > # TODO: for Ron: once the appropriate value is placed in TP > # constructor, see if we should eliminate this parameter from > # description.py. > 'maxSegmentsPerCell': 128, > > # Initial Permanence > # TODO: need better explanation > 'initialPerm': 0.21, > > # Permanence Increment > 'permanenceInc': 0.1, > > # Permanence Decrement > # If set to None, will automatically default to tpPermanenceInc > # value. > 'permanenceDec' : 0.1, > > 'globalDecay': 0.0, > > 'maxAge': 0, > > # Minimum number of active synapses for a segment to be > considered > # during search for the best-matching segments. > # None=use default > # Replaces: tpMinThreshold > 'minThreshold': 9, > > # Segment activation threshold. > # A segment is active if it has >= tpSegmentActivationThreshold > # connected synapses that are active due to infActiveState > # None=use default > # Replaces: tpActivationThreshold > 'activationThreshold': 12, > > 'outputType': 'normal', > > # "Pay Attention Mode" length. This tells the TP 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': 3, > }, > > # Don't create the classifier since we don't need predictions. > 'clEnable': False, > 'clParams': None, > > 'anomalyParams': { u'anomalyCacheRecords': None, > u'autoDetectThreshold': None, > u'autoDetectWaitRecords': 2184}, > > 'trainSPNetOnlyIfRequested': False, > }, > } > > On Tue, Oct 21, 2014 at 2:10 PM, Subutai Ahmad <[email protected]> > wrote: > >> Hi Nicolas, >> >> Yeah, 913 years might be too long! Can you email the model params file >> you are using? I could look at that and try to suggest some changes. >> >> However, 10 msecs per data point is not too bad for the HTM. It might be >> hard to get too much faster! >> >> --Subutai >> >> On Tue, Oct 21, 2014 at 1:45 PM, Nicolas Thiébaud <[email protected]> >> wrote: >> >>> Hi, >>> >>> What are the parameters we can tune to train and predict faster. I am >>> seeing ingestion speed without learning at 100 scalar points points per >>> second. >>> >>> If I downsample my input data (EEG) to 100Hz which is a minimum >>> acceptable and using a 16 core amazon instance, I would need 913 years (not >>> corrected for Moore's law). >>> >>> Which model params should I try to tune? >>> >>> Thank you for the help, >>> >>> Nicolas. >>> >> >> >
