Marcus Lewis <mrcslws@gmail.com>: Author Summary

Builds triggered by Marcus Lewis <mrcslws@gmail.com>

Builds triggered by an author are those builds which contains changes committed by the author.
7
1 (14%)
6 (86%)

Breakages and fixes

Broken means the build has failed but the previous build was successful.
Fixed means that the build was successful but the previous build has failed.
0 (0% of all builds triggered)
1 (14% of all builds triggered)
1
Build Completed Code commits Tests
RES › DW › #270 1 week ago
Merge pull request #816 from mrcslws/multi-column-location-inference
New experiment: Multi-column location inference
New experiment: Multi-column location inference
This uses a new algorithm for calculating the allocentric location:
cortical columns recall all of the allocentric locations where they've
ever sensed an input, and then the cortical columns vote on the body's
allocentric location. The body's allocentric location is calculated
from the sensor's allocentric location and the sensor's egocentric
location. The location is represented by an array of "location
modules" that are inspired by grid cell modules.

This is similar in principle to my previous algorithm that infers the
allocentric location by performing path integration on unions of
locations. The difference here is that this new solution doesn't
require movement. It still benefits from movement, but it can also
infer locations with multiple cortical columns with one "touch".

(I'll probably do more writing on this topic.)
Denote the hovered sensor with color, not size
Bugfix: draw the firing fields in the right place
197 passed
NUP › VIS › #127 1 week ago
Bugfix: draw the firing fields in the right place
New experiment: Multi-column location inference
This uses a new algorithm for calculating the allocentric location:
cortical columns recall all of the allocentric locations where they've
ever sensed an input, and then the cortical columns vote on the body's
allocentric location. The body's allocentric location is calculated
from the sensor's allocentric location and the sensor's egocentric
location. The location is represented by an array of "location
modules" that are inspired by grid cell modules.

This is similar in principle to my previous algorithm that infers the
allocentric location by performing path integration on unions of
locations. The difference here is that this new solution doesn't
require movement. It still benefits from movement, but it can also
infer locations with multiple cortical columns with one "touch".

(I'll probably do more writing on this topic.)
Merge pull request #816 from mrcslws/multi-column-location-inference
New experiment: Multi-column location inference
Denote the hovered sensor with color, not size
Testless build
RES › DW › #256 3 weeks ago
Make discrepancy math easier to read, fix a typo
This code had a typo:
  len(remFFcells > discrepancy)
rather than
  len(remFFcells) > discrepancy

Luckily, it wasn't causing issues.

Thanks @vincentliuk for finding this.
Perf: Don't use random sample when not necessary
Math: Make order of operations more obvious
Merge branch 'master' into column-pooler-discrepancy-math
Merge pull request #809 from mrcslws/column-pooler-discrepancy-math
Make discrepancy math easier to read, fix a typo
197 passed
NUP › VIS › #122 3 weeks ago
Make discrepancy math easier to read, fix a typo
This code had a typo:
  len(remFFcells > discrepancy)
rather than
  len(remFFcells) > discrepancy

Luckily, it wasn't causing issues.

Thanks @vincentliuk for finding this.
Math: Make order of operations more obvious
Perf: Don't use random sample when not necessary
Merge pull request #809 from mrcslws/column-pooler-discrepancy-math
Make discrepancy math easier to read, fix a typo
Merge branch 'master' into column-pooler-discrepancy-math
Testless build
RES › L2L4 › #48 11 months ago
Merge pull request #659 from mrcslws/SetMemory
Add diagrams to SetMemory notebook
Merge pull request #665 from mrcslws/column-pooler-test-reset
Some Column Pooler tests weren't testing the proximal connections
Some Column Pooler tests were not testing the proximal connections
Because we weren't performing a reset after learning, some column
pooler tests would pass even if no proximal connections were formed.
If they just kept the previous active cells active, they would pass.
In some cases, this is what was happening (because the proximal
connections were below the connectedPermanence).
Add diagrams
Merge pull request #658 from mrcslws/SetMemory
Add "SetMemory_similar_objects" notebook
Update SetMemory diagrams
Add "SetMemory_similar_objects" notebook
Merge pull request #661 from mrcslws/SetMemory
Update SetMemory diagrams
Testless build
RES › L2L4 › #40 1 year ago
Update experiment to touch each object for 3 time steps
Add "similar objects" experiment notebook
Bring back a 'compute' method for ColumnPooler
Update tests to use ColumnPooler's compute method
Merge pull request #655 from mrcslws/similar-objects
Re-add ColumnPooler "compute", add "similar objects" notebook
Clarification
Merge pull request #3383 from mrcslws/typo-cleanup
Fix a typo in TM, an unused variable in Connections
Fix a typo and an unused variable
Testless build
UN › UN › #264 1 year ago
y-scaling: handle case where min == max
The y scale was broken when there was only one point on-screen.
Testless build
Build Completed Code commits Tests
Build Completed Code commits Tests
RES › L2L4 › #40 1 year ago
Update experiment to touch each object for 3 time steps
Add "similar objects" experiment notebook
Bring back a 'compute' method for ColumnPooler
Update tests to use ColumnPooler's compute method
Merge pull request #655 from mrcslws/similar-objects
Re-add ColumnPooler "compute", add "similar objects" notebook
Clarification
Merge pull request #3383 from mrcslws/typo-cleanup
Fix a typo in TM, an unused variable in Connections
Fix a typo and an unused variable
Testless build