Params File

Overview

The params file is the recipe for every PetaVision run you cook up. It describes the [Hypercol], [Layers], and [Connections] in the neural network you create including how you want PetaVision to run and save data and checkpoints. By customizing your params file, you can make a stack, and MLP, even a hierarchy of V1 layers ... the only limit to your imagination is the size of the computer you plan to run on.

---

[HyPerCol]

Described at the top of the params file, the HyPerCol is the largest container for the objects (layers and connections) of your run. Here's an example of a HyPerCol from a params file:

HyPerCol "column" = {
    startTime                           = 0;
    dt                                  = 1;
    dtAdaptFlag                         = true;
    dtScaleMax                          = 5;
    dtScaleMin                          = 0.25;
    dtChangeMax                         = 0.1;
    dtChangeMin                         = 0;
    dtMinToleratedTimeScale             = 0.0001;
    stopTime                            = 5e+06;
    progressInterval                    = 1000;
    writeProgressToErr                  = true;
    verifyWrites                        = true;
    outputPath                          = "/nh/compneuro/Data/sparseFuture/output/updownSample";
    printParamsFilename                 = "pv.params";
    randomSeed                          = 1234567890;
    nx                                  = 480;
    ny                                  = 480;
    filenamesContainLayerNames          = 1;
    filenamesContainConnectionNames     = 1;
    initializeFromCheckpointDir         = "";
    checkpointWrite                     = true;
    checkpointWriteDir                  = "/nh/compneuro/Data/sparseFuture/output/updownSample/Checkpoints";
    checkpointWriteTriggerMode          = "step";
    checkpointWriteStepInterval         = 50000;
    deleteOlderCheckpoints              = false;
    suppressNonplasticCheckpoints       = false;
    writeTimescales                     = true;
    errorOnNotANumber                   = false;
};

It's easiest to think about the HyPerCol in terms of two broad categories:

Timing

File I/O

---

[Layers]

PetaVision creates artificial neural networks and organizes the neurons in layers. Layers are described in dimensions relative to the HyPerCol and can ha

---

[Connections]