Temporal Memory is actually behaving as expected. If you look at the log below,
multiple predictions correspond to output of TP (predictive states), best
prediction to classifier output.
The TP is perfectly predicting the next character. The classifier on the other
hand is almost always off (even with 1000 training passes).
Training over sequence ['B', 'A', 'B', 'D', 'D', 'D']
Training over sequence ['C', 'A', 'B', 'D', 'E', 'D']
Testing over sequence ['B', 'A', 'B', 'D', 'D', 'D']
Actual value / Multiple Predictions / Best Prediction = B / none / A
Actual value / Multiple Predictions / Best Prediction = A / A / B
Actual value / Multiple Predictions / Best Prediction = B / B / D
Actual value / Multiple Predictions / Best Prediction = D / D / D
Actual value / Multiple Predictions / Best Prediction = D / D / D
Actual value / Multiple Predictions / Best Prediction = D / D / E
Testing over sequence ['C', 'A', 'B', 'D', 'E', 'D']
Actual value / Multiple Predictions / Best Prediction = C / none / A
Actual value / Multiple Predictions / Best Prediction = A / A / B
Actual value / Multiple Predictions / Best Prediction = B / B / D
Actual value / Multiple Predictions / Best Prediction = D / D / E
Actual value / Multiple Predictions / Best Prediction = E / E / D
Actual value / Multiple Predictions / Best Prediction = D / D / C
This is how the classifier is being trained with alpha = 0.001 and avalpha =
0.3:
active_idx = list(tp.getActiveState().nonzero()[0]) % Current
active cell states
classifier.compute(record_num, active_idx,
{'bucketIdx':ord(s[j])-ord('A'),'actValue':s[j]}, True, True) % s[j] is the
string of the current character, ord(s[j])-ord('A’) finds its bucket index
record_num += 1 % record num is incremented for the next time step
Inference uses the same code with learning turned OFF.
During learning, I tried providing it with the active state from the previous
step thinking that learning required time shifted inputs to create the right
correlations but it seems that is handled by the bit history class of the
classifier.
Hopefully, this gives you a better idea of whether I’m using the classifier
class correctly or not.
best,
Nick
On Sep 12, 2014, at 12:50 AM, Chetan Surpur <[email protected]> wrote:
> Hi Nick,
>
> I'm not sure what's wrong. The first thing I would check is temporal memory
> parameters.
>
> - Chetan
>
>
> On Sun, Sep 7, 2014 at 2:04 PM, Nicholas Mitri <[email protected]> wrote:
>
> Hey all,
>
> I m bumping this thread for the after weekend activity and would like to add
> on some more inquiries.
> I modified hello_tp.py and added a sequence generator:
>
> generate_seqs(items, seq_num, seq_length, replacement, reproduceable)
>
> e.g. generate_seqs(‘ABCDE', 5, 4, True, True) produces the sequences seen in
> the log appended below.
> The log shows the generated sequences and the predictions being made.
> The first prediction (multiple) shows all the possible character prediction
> i.e. those with sufficient overlap with the predictive state of the network.
> The second one is based on the most likely character as decided by the CLA
> classifier which I imported into hello_tp.py.
>
> Accuracy measures are shown at the end of the log.
> As you can tell, the classifier is performing horribly. I’m not sure if I’m
> not using it correctly so I’ve attached the modified py file.
> I’d appreciate any comments as to why the classifier is failing to predict so
> often.
>
> Thanks,
> Nick
>
> <main.py>
>
> ———— STDOUT ————
>
> Training over sequence ('E', 'C', 'E', 'C', 'C', 'E')
> Training over sequence ('C', 'A', 'A', 'C', 'E', 'A')
> Training over sequence ('B', 'C', 'D', 'E', 'A', 'D')
> Training over sequence ('E', 'B', 'C', 'E', 'C', 'A')
> Training over sequence ('E', 'C', 'B', 'C', 'E', 'C')
>
> Testing over sequence ('E', 'C', 'E', 'C', 'C', 'E')
> Actual value / Multiple Predictions / Best Prediction = E / none / C
> Actual value / Multiple Predictions / Best Prediction = C / BC / E
> Actual value / Multiple Predictions / Best Prediction = E / BE / C
> Actual value / Multiple Predictions / Best Prediction = C / C / C
> Actual value / Multiple Predictions / Best Prediction = C / C / E
> Actual value / Multiple Predictions / Best Prediction = E / E / C
>
> Testing over sequence ('C', 'A', 'A', 'C', 'E', 'A')
> Actual value / Multiple Predictions / Best Prediction = C / none / E
> Actual value / Multiple Predictions / Best Prediction = A / BE / E
> Actual value / Multiple Predictions / Best Prediction = A / A / C
> Actual value / Multiple Predictions / Best Prediction = C / C / E
> Actual value / Multiple Predictions / Best Prediction = E / E / A
> Actual value / Multiple Predictions / Best Prediction = A / A / C
>
> Testing over sequence ('B', 'C', 'D', 'E', 'A', 'D')
> Actual value / Multiple Predictions / Best Prediction = B / none / C
> Actual value / Multiple Predictions / Best Prediction = C / C / D
> Actual value / Multiple Predictions / Best Prediction = D / D / E
> Actual value / Multiple Predictions / Best Prediction = E / E / A
> Actual value / Multiple Predictions / Best Prediction = A / A / D
> Actual value / Multiple Predictions / Best Prediction = D / D / B
>
> Testing over sequence ('E', 'B', 'C', 'E', 'C', 'A')
> Actual value / Multiple Predictions / Best Prediction = E / none / C
> Actual value / Multiple Predictions / Best Prediction = B / BC / C
> Actual value / Multiple Predictions / Best Prediction = C / C / E
> Actual value / Multiple Predictions / Best Prediction = E / E / C
> Actual value / Multiple Predictions / Best Prediction = C / C / A
> Actual value / Multiple Predictions / Best Prediction = A / A / E
>
> Testing over sequence ('E', 'C', 'B', 'C', 'E', 'C')
> Actual value / Multiple Predictions / Best Prediction = E / none / C
> Actual value / Multiple Predictions / Best Prediction = C / BC / E
> Actual value / Multiple Predictions / Best Prediction = B / BE / C
> Actual value / Multiple Predictions / Best Prediction = C / C / E
> Actual value / Multiple Predictions / Best Prediction = E / E / C
> Actual value / Multiple Predictions / Best Prediction = C / C / E
>
>
> RESULTS:
> --------
> We use two evaluation metrics to quantify the performance of the TP sequence
> learning.
> The first is based on multiple predictions being equally weighted i.e.
> without resorting to the CLA Classifier. Here, it is enough for the actual
> value to be among the multiple predictions being made.
> The second and more strict metric leverages the CLA Classifier and compares
> the actual value against the most likely prediction.
> Finally, we provide average prediction scores as an indicater of the
> prediction accuraccy on the character level. Each corrent prediction
> increments the score by 1/(number of predictions made).
> For example, if prediction is ABC and the actual value is A, the prediction
> score is incremented by 1/3.
>
> Prediction Accuracy from metric 1 = 80.00% with an average score of 3.80/5.00
> Prediction Accuracy from metric 2 = 0.00% with an average score of 0.00/5.00
>
>
> On Sep 4, 2014, at 11:37 PM, Nicholas Mitri <[email protected]> wrote:
>
>> Hey all,
>>
>> I’d like to dedicate this thread for discussing some TP implementation and
>> practical questions, namely those associated with the introductory file to
>> the TP, hello-tp.py.
>>
>> Below is the print out of a TP with 50 columns, 1 cell per column being
>> trained as described in the py file for 2 iterations on the sequence
>> A->B->C->D->E. Each pattern is fed directly into the TP as an active network
>> state.
>>
>> I’ve been playing around with the configurations and have a few questions.
>>
>> 1) Why are there no predictive states being seen for the first training pass
>> (i.e. seeing the entire sequence once)? Even if activationThreshold and
>> minThreshold are set sufficiently low to make segments sensitive, no lateral
>> activation happens. Are cells initialized with no segments?
>> 2) if segments are created during initialization, how is their connectivity
>> to the cells of the region configured? How are permanence values allocated?
>> Same as proximal synapses in the TP?
>> 3) in the second training pass, we go from no predictive cells to perfectly
>> predictive cells associated with the next character. I would typically
>> expect the network to show scattered predictive cells before it hones in on
>> the right prediction (consecutive 10 on-bits in this example). Why the
>> abrupt shift in predictive behavior? Is this related to
>> getBestMatchingCell()?
>> 4) finally, the printCells() function outputs the following. Can you please
>> explain what each entry means?
>>
>> Column 41 Cell 0 : 1 segment(s)
>> Seg #0 ID:41 True 0.2000000 ( 3/3 ) 0 [30,0]1.00 [31,0]1.00
>> [32,0]1.00 [33,0]1.00 [34,0]1.00 [35,0]1.00 [36,0]1.00 [37,0]1.00 [38,0]1.00
>> [39,0]1.00
>>
>> Thanks,
>> Nick
>> ————————————— PRINT OUT———————————— —————
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 1111111111 0000000000 0000000000 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 1111111111 0000000000 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 1111111111 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 0000000000 1111111111 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 0000000000 0000000000 1111111111
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> ############ Training Pass #1 Complete ############
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 1111111111 0000000000 0000000000 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 1111111111 0000000000 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 1111111111 0000000000 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 1111111111 0000000000 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 1111111111 0000000000 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 1111111111 0000000000
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 0000000000 1111111111 0000000000
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 1111111111
>>
>> All the active and predicted cells:
>>
>> Inference Active state
>> 0000000000 0000000000 0000000000 0000000000 1111111111
>> Inference Predicted state
>> 0000000000 0000000000 0000000000 0000000000 0000000000
>>
>> ############ Training Pass #2 Complete ############
>
>
