tl; dr 如何使用pySpark比較行的相似性?pySpark Columns相似性問題
我有一個numpy的陣列,我想每一行的相似之處彼此比較
print (pdArray)
#[[ 0. 1. 0. ..., 0. 0. 0.]
# [ 0. 0. 3. ..., 0. 0. 0.]
# [ 0. 0. 0. ..., 0. 0. 7.]
# ...,
# [ 5. 0. 0. ..., 0. 1. 0.]
# [ 0. 6. 0. ..., 0. 0. 3.]
# [ 0. 0. 0. ..., 2. 0. 0.]]
使用SciPy的我可以計算餘弦相似之處遵循...
pyspark.__version__
# '2.2.0'
from sklearn.metrics.pairwise import cosine_similarity
similarities = cosine_similarity(pdArray)
similarities.shape
# (475, 475)
print(similarities)
array([[ 1.00000000e+00, 1.52204908e-03, 8.71545594e-02, ...,
3.97681174e-04, 7.02593036e-04, 9.90472253e-04],
[ 1.52204908e-03, 1.00000000e+00, 3.96760121e-04, ...,
4.04724413e-03, 3.65324300e-03, 5.63519735e-04],
[ 8.71545594e-02, 3.96760121e-04, 1.00000000e+00, ...,
2.62367141e-04, 1.87878869e-03, 8.63876439e-06],
...,
[ 3.97681174e-04, 4.04724413e-03, 2.62367141e-04, ...,
1.00000000e+00, 8.05217639e-01, 2.69724702e-03],
[ 7.02593036e-04, 3.65324300e-03, 1.87878869e-03, ...,
8.05217639e-01, 1.00000000e+00, 3.00229809e-03],
[ 9.90472253e-04, 5.63519735e-04, 8.63876439e-06, ...,
2.69724702e-03, 3.00229809e-03, 1.00000000e+00]])
由於我正在尋找擴大到比我原來的(475行)矩陣更大的集,我正在通過pySpark使用Spark觀看
from pyspark.mllib.linalg.distributed import RowMatrix
#load data into spark
tempSpark = sc.parallelize(pdArray)
mat = RowMatrix(tempSpark)
# Calculate exact similarities
exact = mat.columnSimilarities()
exact.entries.first()
# MatrixEntry(128, 211, 0.004969676943490767)
# Now when I get the data out I do the following...
# Convert to a RowMatrix.
rowMat = approx.toRowMatrix()
t_3 = rowMat.rows.collect()
a_3 = np.array([(x.toArray()) for x in t_3])
a_3.shape
# (488, 749)
正如你所看到的,數據的形狀是a)不再是方形的(它應該是和b)的尺寸與原始行數不匹配......現在它確實匹配(在部分_中的特徵數量在每一行(len(pdArray [0])= 749),但我不知道488是從哪裏來的
749的存在讓我覺得我需要先調換我的數據。那是對的嗎?
最後,如果是這種情況,爲什麼尺寸不是(749,749)?
稀疏向量爲此顯示多少行rowMat.rows.collect()? – Suresh