Write numpy arrays to lmdb

Question

I'm trying to write some numpy arrays in python to lmdb:

import numpy as np
import lmdb

def write_lmdb(filename):
    lmdb_env = lmdb.open(filename, map_size=int(1e9))
    lmdb_txn = lmdb_env.begin(write=True)

    X= np.array([[1.0, 0.0], [0.1, 2.0]])
    y= np.array([1.4, 2.1])

    #Put first pair of arrays
    lmdb_txn.put('X', X)
    lmdb_txn.put('y', y)

    #Put second pair of arrays
    lmdb_txn.put('X', X+1.6)
    lmdb_txn.put('y', y+1.2)

def read_lmdb(filename):
    lmdb_env = lmdb.open(filename)
    lmdb_txn = lmdb_env.begin()
    lmdb_cursor = lmdb_txn.cursor()
    for key, value in lmdb_cursor:
        print type(key)
        print type(value)

        print key
        print value

write_lmdb('temp.db')
read_lmdb('temp.db')

but read_lmdb prints nothing, what is the proper way to write numpy arrays to lmdb?

Update: Based on @frankyjuang answer I manage to do it, howewer not in very elegant way: multidimensional array lose it's shape, each array should have it's own name.

import numpy as np
import lmdb

def write_lmdb(filename):
    print 'Write lmdb'

    lmdb_env = lmdb.open(filename, map_size=int(1e9))

    n_samples= 2
    X= (255*np.random.rand(n_samples,3,4,3)).astype(np.uint8)
    y= np.random.rand(n_samples).astype(np.float32)

    for i in range(n_samples):
        with lmdb_env.begin(write=True) as lmdb_txn:
            lmdb_txn.put('X_'+str(i), X)
            lmdb_txn.put('y_'+str(i), y)

            print 'X:',X
            print 'y:',y

def read_lmdb(filename):
    print 'Read lmdb'

    lmdb_env = lmdb.open(filename)
    lmdb_txn = lmdb_env.begin()
    lmdb_cursor = lmdb_txn.cursor()

    n_samples=0
    with lmdb_env.begin() as lmdb_txn:
        with lmdb_txn.cursor() as lmdb_cursor:
            for key, value in lmdb_cursor:  
                print key
                if('X' in key):
                    print np.fromstring(value, dtype=np.uint8)
                if('y' in key):
                    print np.fromstring(value, dtype=np.float32)

                n_samples=n_samples+1

    print 'n_samples',n_samples

write_lmdb('temp.db')
read_lmdb('temp.db')

Test script output should be something like:

Write lmdb
X: [[[[ 48 224 119]
   [ 76  87 174]
   [ 14  88 183]
   [ 76 234  56]]

  [[234 223  65]
   [ 63  85 175]
   [184 252 125]
   [100   7 225]]

  [[134 159  41]
   [  2 146 221]
   [ 99  74 225]
   [169  57  59]]]


 [[[100 202   3]
   [ 88 204 131]
   [ 96 238 243]
   [103  58  30]]

  [[157 125 107]
   [238 207  99]
   [102 220  64]
   [ 27 240  33]]

  [[ 74  93 131]
   [107  88 206]
   [ 55  86  35]
   [212 235 187]]]]
y: [ 0.80826157  0.01407595]
X: [[[[ 48 224 119]
   [ 76  87 174]
   [ 14  88 183]
   [ 76 234  56]]

  [[234 223  65]
   [ 63  85 175]
   [184 252 125]
   [100   7 225]]

  [[134 159  41]
   [  2 146 221]
   [ 99  74 225]
   [169  57  59]]]


 [[[100 202   3]
   [ 88 204 131]
   [ 96 238 243]
   [103  58  30]]

  [[157 125 107]
   [238 207  99]
   [102 220  64]
   [ 27 240  33]]

  [[ 74  93 131]
   [107  88 206]
   [ 55  86  35]
   [212 235 187]]]]
y: [ 0.80826157  0.01407595]
Read lmdb
X_0
[ 48 224 119  76  87 174  14  88 183  76 234  56 234 223  65  63  85 175
 184 252 125 100   7 225 134 159  41   2 146 221  99  74 225 169  57  59
 100 202   3  88 204 131  96 238 243 103  58  30 157 125 107 238 207  99
 102 220  64  27 240  33  74  93 131 107  88 206  55  86  35 212 235 187]
X_1
[ 48 224 119  76  87 174  14  88 183  76 234  56 234 223  65  63  85 175
 184 252 125 100   7 225 134 159  41   2 146 221  99  74 225 169  57  59
 100 202   3  88 204 131  96 238 243 103  58  30 157 125 107 238 207  99
 102 220  64  27 240  33  74  93 131 107  88 206  55  86  35 212 235 187]
y_0
[ 0.80826157  0.01407595]
y_1
[ 0.80826157  0.01407595]
n_samples 4

Answer 1

Wrap your transactions under with. And remember to convert the value from bytes (string) back to numpy array using np.fromstring.

To be honest, it is not a good idea to store numpy array in lmdb since conversion from array to bytes back to array will lose some informations (ex. shape). You can try storing a dict of numpy arrays using pickle.

def write_lmdb(filename):
    ...
    with lmdb_env.begin(write=True) as lmdb_txn:
        ...

def read_lmdb(filename):
    ...
    with lmdb_env.begin() as lmdb_txn:
        with lmdb_txn.cursor() as lmdb_cursor:
            ...
            print np.fromstring(value, dtype=np.float64)