We are continuing this conversation on Gitter:
https://gitter.im/numenta/public/archives/2015/11/02
---------
Matt Taylor
OS Community Flag-Bearer
Numenta


On Mon, Nov 2, 2015 at 12:08 PM, Cas <[email protected]> wrote:
> hey Matt,
>
> I don't have access to my machine right now, but I've checked out the
> metrics data table in a db gui and I recall seeing all null values in the
> anomaly scores columns. I think the total amount of recorded data points is
> around 230000 (it's the default set from the tutorial). I'll double check
> tomorrow morning, it would definitely be silly if that was it!
>
> Meanwhile I've tried to track down the code responsible for sending the data
> into HTM Engine. The fact that I'm unfamiliar with python complicates this,
> though. I'm thinking maybe anomaly_service.py or a file in model_swapper
> could tell me more? In terms of configs I was thinking model checkpoints
> might not be working right, even though I'm not sure what that does :)
>
>
>
> Met vriendelijke groet,
>
> Casper Rooker
> [email protected]
>
> On Mon, Nov 2, 2015 at 4:30 PM, Matthew Taylor <[email protected]> wrote:
>>
>> Cas,
>>
>> You won't see an anomaly score out of the HTM Engine until it has seen
>> 500 data points. Has it seen that much data yet?
>>
>> ---------
>> Matt Taylor
>> OS Community Flag-Bearer
>> Numenta
>>
>>
>> On Mon, Nov 2, 2015 at 4:25 AM, Cas <[email protected]> wrote:
>> > As I said I'm trying out the HTM engine traffic tutorial where all the
>> > model
>> > params are generated automatically when the client starts offering data
>> > to
>> > the HTTP API.
>> >
>> > Now, I don't know exactly how to interpret the metric records, but I can
>> > at
>> > least tell that the inferencetype is what it should be.
>> >
>> > So I've concluded that part of the tutorial that runs the data through
>> > HTM
>> > Engine probably isn't doing it's job. The logs in supervisord tell me
>> > all
>> > the services are working fine (though I had to manually open the port
>> > for
>> > htmengine:model_scheduler). I would assume that the traffic tutorial is
>> > configured to do postprocessing, based on the HTM engine's README.
>> >
>> > It's probably a silly oversight on my part, but right now I just dont
>> > know
>> > what to make of this. Hope someone can help out!
>> >
>> > Kind regards,
>> >
>> > Casper Rooker
>> > [email protected]
>> >
>> > P.S.:
>> >
>> > One such model params record in the metrics table:
>> >
>> > {
>> >   "inferenceArgs": {
>> >     "predictionSteps": [
>> >       1
>> >     ],
>> >     "predictedField": "c1",
>> >     "inputPredictedField": "auto"
>> >   },
>> >   "modelConfig": {
>> >     "aggregationInfo": {
>> >       "seconds": 0,
>> >       "fields": [],
>> >       "months": 0,
>> >       "days": 0,
>> >       "years": 0,
>> >       "hours": 0,
>> >       "microseconds": 0,
>> >       "weeks": 0,
>> >       "minutes": 0,
>> >       "milliseconds": 0
>> >     },
>> >     "model": "CLA",
>> >     "version": 1,
>> >     "predictAheadTime": null,
>> >     "modelParams": {
>> >       "sensorParams": {
>> >         "verbosity": 0,
>> >         "encoders": {
>> >           "c0_dayOfWeek": null,
>> >           "c0_timeOfDay": {
>> >             "fieldname": "c0",
>> >             "timeOfDay": [
>> >               21,
>> >               9.49122334747737
>> >             ],
>> >             "type": "DateEncoder",
>> >             "name": "c0"
>> >           },
>> >           "c1": {
>> >             "name": "c1",
>> >             "resolution": 0.7017543859649122,
>> >             "seed": 42,
>> >             "fieldname": "c1",
>> >             "type": "RandomDistributedScalarEncoder"
>> >           },
>> >           "c0_weekend": null
>> >         },
>> >         "sensorAutoReset": null
>> >       },
>> >       "clEnable": false,
>> >       "spParams": {
>> >         "columnCount": 2048,
>> >         "spVerbosity": 0,
>> >         "maxBoost": 1.0,
>> >         "spatialImp": "cpp",
>> >         "inputWidth": 0,
>> >         "synPermInactiveDec": 0.0005,
>> >         "synPermConnected": 0.1,
>> >         "synPermActiveInc": 0.0015,
>> >         "seed": 1956,
>> >         "numActiveColumnsPerInhArea": 40,
>> >         "globalInhibition": 1,
>> >         "potentialPct": 0.8
>> >       },
>> >       "trainSPNetOnlyIfRequested": false,
>> >       "clParams": {
>> >         "alpha": 0.035828933612158,
>> >         "clVerbosity": 0,
>> >         "steps": "1",
>> >         "regionName": "CLAClassifierRegion"
>> >       },
>> >       "tpParams": {
>> >         "columnCount": 2048,
>> >         "activationThreshold": 13,
>> >         "pamLength": 3,
>> >         "cellsPerColumn": 32,
>> >         "permanenceInc": 0.1,
>> >         "minThreshold": 10,
>> >         "verbosity": 0,
>> >         "maxSynapsesPerSegment": 32,
>> >         "outputType": "normal",
>> >         "globalDecay": 0.0,
>> >         "initialPerm": 0.21,
>> >         "permanenceDec": 0.1,
>> >         "seed": 1960,
>> >         "maxAge": 0,
>> >         "newSynapseCount": 20,
>> >         "maxSegmentsPerCell": 128,
>> >         "temporalImp": "cpp",
>> >         "inputWidth": 2048
>> >       },
>> >       "anomalyParams": {
>> >         "anomalyCacheRecords": null,
>> >         "autoDetectThreshold": null,
>> >         "autoDetectWaitRecords": 5030
>> >       },
>> >       "spEnable": true,
>> >       "inferenceType": "TemporalAnomaly",
>> >       "tpEnable": true
>> >     }
>> >   },
>> >   "inputRecordSchema": [
>> >     [
>> >       "c0",
>> >       "datetime",
>> >       "T"
>> >     ],
>> >     [
>> >       "c1",
>> >       "float",
>> >       ""
>> >     ]
>> >   ]
>> > }
>> >
>> > On Mon, Nov 2, 2015 at 12:14 PM, Wakan Tanka <[email protected]> wrote:
>> >>
>> >> On 11/02/2015 11:31 AM, Cas wrote:
>> >>>
>> >>> Hello NuPIC,
>> >>>
>> >>> I was trying out the HTM engine traffic tutorial today and I got it
>> >>> running. Unfortunately the anomaly score is 'none' for all data
>> >>> points.
>> >>>
>> >>> Do you have any suggestions on how to troubleshoot this?
>> >>>
>> >>> I know all the services are running, from the looks of the supervisord
>> >>> interface. I'd like to see how the data points are being offered to
>> >>> the
>> >>> HTM engine for starters, I'm just not sure how to do that.
>> >>>
>> >>> Kind regards,
>> >>>
>> >>> Casper Rooker
>> >>> [email protected] <mailto:[email protected]>
>> >>
>> >>
>> >> Hello Casper,
>> >>
>> >> Be sure that you've set "inferenceType": "TemporalAnomaly" if you have
>> >> "inferenceType": "TemporalMultiStep" then you are unable to get anomaly
>> >> score just predictions.
>> >>
>> >> regards
>> >>
>> >> Wakan Tanka
>> >>
>> >
>>
>

Reply via email to