Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
tree: b2a4f8bf39

Fetching latest commit…

Cannot retrieve the latest commit at this time

..
Failed to load latest commit information.
README
apply-odd-kernels
apply-svm-to-exp
codex
compute-roc
d-prime
draw-bounding-box
draw-s2-bounding-box
exp-get-results
exp-imprint-prototypes
exp-make-splits
exp-run
exp-svm-crossval
exp-svm-for-splits
exp-svm-multi-for-splits
exp-transform-corpus
exp2orange
find-smallest-input-size
format-glab-results
get-num-cores
gjob
glimpse
glimpse-cluster-wrapper
imprint-prototypes
imprint-random-prototypes
learn-odd-kernels
make-random-kernels
map-layer-size
map-region
multi-launch
np-concat
np-generate
np-shape
np-split
np-stats
np-to-svm
optimize-svm-param
parse-stack-trace.py
petacat-apply-c1
petacat-apply-c2
petacat-learn-c1
petacat-learn-c2
plot
pure_python_transform
remote-job-manager
remote-job-manager-results
scaling-comparison
show-array
show-odd-kernels
svm
svm-crossval
svm-evaluate
svm-to-np
transform
transform-dir
transform-single-scale
voc-get-annotations
voc-get-layerwise-crops
voc-make-crops

README

== Basic Utilities ==

codex -- transform data files from one format to another. Formats include:
  p: (p)ython-native, binary-encoded format (using the pickle module)
  c: (c)omma-seperated text format
  s: free-form text (s)ummary
np-concat -- Join two arrays.
np-generate -- Evaluate a set of arbitrary Python expressions, writing results
               to the terminal or a file.
np-shape -- Get the shape of an array.
np-split -- Partition an array into a number of segments.
np-stats -- Compute statistics (min, max, mean, std) of values in an array.
show-array -- Convert an N-D array into a greyscale image.

== Experiment Batch Scripts ==
These scripts run a transformation on multiple files at once, and apply an SVM
to the results.

exp-make-splits -- Create training and testing splits for an experiment.
exp-run -- Main experiment control script (see example setup script at
           rc/example-experiment.sh).
exp-svm-crossval -- Compute cross-validated SVM accuracy on transformed images.
exp-svm-for-splits -- Run SVM on transformed images with existing train/test
                      splits.
exp-transform-corpus -- Transform a set of images, and optionally apply an SVM
                        to the results.
get-num-cores -- Tries to determine the number of available processor cores on
                 the host machine.
multi-launch -- Launch a given command N times in background processes, waiting
                for results.

== Hierarchical Models ==
These scripts perform one step of a complete experiment.

draw-bounding-box -- Visualize a bounding box given by points in layer-dependent
                     coordinate system.
draw-s2-bounding-box --
imprint-prototypes -- Generate S2 prototypes by "imprinting" from specific
                      regions of C1 activity computed from a given image.
imprint-random-prototypes -- Generate S2 prototypes by randomly selecting
                             regions of C1 activity computed from an image.
make-random-kernels -- Create a set of randomly-generated S-unit kernels.
transform -- Apply an HMAX-like hierarchy to a single image.
transform-dir -- Apply 'transform' to all images in a list.

== SVM Utilities ==
These scripts apply a linear SVM to an arbitrary set of feature vectors.

compute-roc -- Compute the ROC curve for some classification results.
d-prime -- Compute the d' measure for some classification results.
np-to-svm -- Write a numpy array in LIBSVM format.
optimize-svm-param -- Searches for the optimal setting of the SVM margin penalty
                      with respect to a given training set
svm -- Apply linear SVM to existing training/testing data. Requires the LIBSVM
       and (optionally) SVM-light packages.
svm-classify -- Apply a trained SVM model to a set of instances.
svm-crossval -- Test an SVM using NxM-way cross-validation.
svm-to-np -- Write a dataset in LIBSVM format as a numpy array.

== Job Control Scripts ==
These scripts control an arbitrary job (e.g., an experiment using exp-run) on a
local or remote host.

local-job-kill -- Stop (not pause) a local job.
local-job-delete -- Remove the job files (not directory) of a local job.
local-job-mkdir -- Make a new directory for a local job.
local-job-start -- Launch a new job on the local machine.
local-job-status -- Get the status of a job running on the local machine, given
                    the path to its directory.

== Petacat Commands ==

petacat-apply-c2 -- Use an previously trained SVM model to classify an image.
petacat-learn-c2 -- Given a set of training images, compute the C2 activity from
                    an HMAX-like model and train an SVM classifier.

== Lateral Kernels ==
apply-odd-kernels -- Apply a set of ODD kernels to the S1 activity of an image.
learn-odd-kernels -- Construct per-class co-occurrence matrices from training
                     images.
show-odd-kernels -- Compute ODD kernel for per-class co-occurrence matrices,
                    and plot all orientation fields.

== Job Control Scripts ==
remote-job-manager -- Manages M jobs on N compute nodes, for M > N.
remote-job-manager-results -- Collect the results from a set of remote jobs.
remote-job-cmd -- Run a command in the remote job directory.
remote-job-is-paused -- Determine if a remote job is currently paused.
remote-job-kill -- Stop (not pause) a job running on a remote compute node.
remote-job-pause -- Pause a job running on a remote compute node.
remote-job-start -- Launch a new job on a remote compute node.
remote-job-status -- Get the status of a job on a remote compute node.
remote-job-unpause -- Resume a paused job on a remote compute node.

== Other ==

map-region -- Convert a range expressed in one layer's coordinates into the
              equivalent range in a lower-layer's coordinates.
map-layer-size -- Compute the image size required to support a model layer
                  (e.g., C2) with at least the given spatial extent (e.g., 3x3
                  units).
Something went wrong with that request. Please try again.