Created
February 12, 2014 16:38
-
-
Save strubell/8959278 to your computer and use it in GitHub Desktop.
maxtrix, vector left multiply vs. left multiply + sum + max
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
def leftMultiplyAndSumAndMax(t: Tensor1, v: DenseTensor1): (DenseTensor1, Double) = { | |
assert(dim1 == t.dim1, "Dimensions don't match: " + dim1 + " " + t.dim1) | |
val myDim2 = dim2 | |
val newT = v.copy | |
val newArray = newT.asArray | |
var max = Double.MinValue | |
var currentVal = 0.0 | |
t match { | |
case t: DenseTensor => | |
val tArr = t.asArray | |
var row = 0 | |
while (row < tArr.length-1) { | |
val v = tArr(row) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
col += 1 | |
} | |
row += 1 | |
} | |
/* Do final iteration separately since it's the only one where we need max logic */ | |
val v = tArr(row) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
currentVal = newArray(col) | |
if(currentVal > max){ | |
max = currentVal | |
} | |
col += 1 | |
} | |
case t: SparseBinaryTensor => | |
val tActiveDomainSize = t.activeDomainSize | |
val tIndices = t._indices | |
var ti = 0 | |
while (ti < tActiveDomainSize-1) { | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += apply(offset + col) | |
currentVal = newArray(col) | |
col += 1 | |
} | |
ti += 1 | |
} | |
/* Do final iteration separately since it's the only one where we need max logic */ | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += apply(offset + col) | |
currentVal = newArray(col) | |
if(currentVal > max) | |
max = currentVal | |
col += 1 | |
} | |
case t: SparseIndexedTensor => | |
val tActiveDomainSize = t.activeDomainSize | |
val tIndices = t._indices | |
val tValues = t._values | |
var ti = 0 | |
while (ti < tActiveDomainSize-1) { | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
val v = tValues(ti) | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
col += 1 | |
} | |
ti += 1 | |
} | |
/* Do final iteration separately since it's the only one where we need max logic */ | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
val v = tValues(ti) | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
currentVal = newArray(col) | |
if(currentVal > max) | |
max = currentVal | |
col += 1 | |
} | |
case _ => | |
throw new Error("tensor type neither dense nor sparse: " + t.getClass.getName) | |
} | |
(newT, max) | |
} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
def leftMultiply(t: Tensor1): Tensor1 = { | |
assert(dim1 == t.dim1, "Dimensions don't match: " + dim1 + " " + t.dim1) | |
val myDim2 = dim2 | |
val newT = new DenseTensor1(dim2) | |
val newArray = newT.asArray | |
t match { | |
case t: DenseTensor => | |
val tArr = t.asArray | |
var row = 0 | |
while (row < tArr.length) { | |
val v = tArr(row) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
col += 1 | |
} | |
row += 1 | |
} | |
case t: SparseBinaryTensor => | |
val tActiveDomainSize = t.activeDomainSize | |
val tIndices = t._indices | |
var ti = 0 | |
while (ti < tActiveDomainSize) { | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += apply(offset + col) | |
col += 1 | |
} | |
ti += 1 | |
} | |
case t: SparseIndexedTensor => | |
val tActiveDomainSize = t.activeDomainSize | |
val tIndices = t._indices | |
val tValues = t._values | |
var ti = 0 | |
while (ti < tActiveDomainSize) { | |
val row = tIndices(ti) | |
val offset = row * myDim2 | |
val v = tValues(ti) | |
var col = 0 | |
while (col < myDim2) { | |
newArray(col) += (apply(offset + col) * v) | |
col += 1 | |
} | |
ti += 1 | |
} | |
case _ => | |
throw new Error("tensor type neither dense nor sparse: " + t.getClass.getName) | |
} | |
newT | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment