Update: generate efanna graph

.gitattributes

@@ -68,3 +68,6 @@ efanna_graph/glove_50k_50.graph filter=lfs diff=lfs merge=lfs -text
 efanna_graph/sift1m.50nn.graph filter=lfs diff=lfs merge=lfs -text
 efanna_graph/glove_50k_50nn.graph filter=lfs diff=lfs merge=lfs -text
 efanna_graph/deep1b_base.10M.50nn.graph filter=lfs diff=lfs merge=lfs -text
+efanna_graph/deep1b_base.10M.400nn.graph filter=lfs diff=lfs merge=lfs -text
+efanna_graph/glove_50k_400nn.graph filter=lfs diff=lfs merge=lfs -text
+efanna_graph/sift1m.200nn.graph filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,2 +1,4 @@
 __pycache__/
 anns/build_efanna_log.txt
+venv/
+.idea/

build_efanna_graph.sh

@@ -1,4 +1,4 @@
 # test_nndescent data_file save_graph K L iter S R
-test_nndescent sift1m/sift_base.fvecs efanna_graph/sift1m.50nn.graph 100 120 10 15 70
-test_nndescent glove/glove_50k_50.fvecs efanna_graph/glove_50k_50nn.graph 100 150 10 15 70
-test_nndescent deep1b/base.10M.fvecs efanna_graph/deep1b_base.10M.50nn.graph 100 150 10 15 70
+test_nndescent sift1m/sift_base.fvecs efanna_graph/sift1m.200nn.graph 200	200	12	10	100
+test_nndescent glove/glove_50k_50.fvecs efanna_graph/glove_50k_400nn.graph 400	420	12	20	200
+test_nndescent deep1b/base.10M.fvecs efanna_graph/deep1b_base.10M.400nn.graph 400	400	12	15	100

efanna_graph/deep1b_base.10M.400nn.graph

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef95d1890525e5e598309f52312cf43c8716c52e77829dff720ab431aa128c5b
+size 16040000000

efanna_graph/glove_50k_400nn.graph

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0fec0e106890fb1fe3062e51dcc9c5ecacf5e204bddb92fdd378ac494f7c10b
+size 80200000

efanna_graph/sift1m.200nn.graph

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10eccb43f62a00331987f8a255194cf2b19c01074d36de42447aadad6536d771
+size 804000000

format_deep1b.py

@@ -1,10 +0,0 @@
-from deep1b.loading import read_fbin
-import numpy as np
-
-# chunk_size = 50000
-dataset = read_fbin('deep1b/base.10M.fbin', chunk_size=None)
-
-np.save('deep1b/base.10M.npy', dataset)
-
-from to_fvecs import data_to_fvecs
-data_to_fvecs(dataset, 'deep1b/base.10M.fvecs')

format_glove.py

@@ -1,10 +0,0 @@
-from glove.loading import load
-import numpy as np
-
-# chunk_size = 50000
-dataset = load('glove/glove_50k_50.txt')
-
-np.save('glove/glove_50k_50.npy', dataset)
-
-from to_fvecs import data_to_fvecs
-data_to_fvecs(dataset, 'glove/glove_50k_50.fvecs')

format_sift1m.py

@@ -1,7 +0,0 @@
-from sift1m.loading import fvecs_read
-import numpy as np
-
-dataset,_,_ = fvecs_read('sift1m/sift_base.fvecs')
-print(f"{dataset.shape}")
-np.save('sift1m/sift_base.npy', dataset)
-

glove/loading.py

@@ -1,4 +0,0 @@
-from gensim.models import KeyedVectors
-
-def load(filename):
-    return KeyedVectors.load_word2vec_format(filename).vectors

process/format_deep1b.py

@@ -0,0 +1,10 @@
+from load import load_deep1b
+import numpy as np
+
+# chunk_size = 50000
+dataset = read_fbin('../deep1b/base.10M.fbin', chunk_size=None)
+
+np.save('../deep1b/base.10M.npy', dataset)
+
+from to_fvecs import data_to_fvecs
+data_to_fvecs(dataset, '../deep1b/base.10M.fvecs')

process/format_glove.py

@@ -0,0 +1,8 @@
+from load import load_glove
+import numpy as np
+
+dataset = load_glove('../glove/glove_50k_50.txt')
+np.save('../glove/glove_50k_50.npy', dataset)
+
+from to_fvecs import data_to_fvecs
+data_to_fvecs(dataset, '../glove/glove_50k_50.fvecs')

process/format_sift1m.py

@@ -0,0 +1,5 @@
+from load import load_sift1m
+import numpy as np
+
+dataset,_,_ = load_sift1m('../sift1m/sift_base.fvecs')
+np.save('../sift1m/sift_base.npy', dataset)

deep1b/loading.py → process/load.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-def read_fbin(filename, start_idx=0, chunk_size=None):
+def load_deep1b(filename, start_idx=0, chunk_size=None):
     """ Read *.fbin file that contains float32 vectors
     Args:
         :param filename (str): path to *.fbin file
@@ -17,8 +17,15 @@ def read_fbin(filename, start_idx=0, chunk_size=None):
                           offset=start_idx * 4 * dim)
     return arr.reshape(nvecs, dim)
 
+def load_glove(filename):
+    from gensim.models import KeyedVectors
+    return KeyedVectors.load_word2vec_format(filename).vectors
 
-if __name__=='__main__':
-# read data
-    chunk_size = 50000
-    dataset = read_fbin('deep1B/base.10M.fbin', chunk_size=chunk_size)
+
+def load_sift1m(fname):
+    data = np.fromfile(fname, dtype=np.float32)
+    dim = data[0].view(np.int32)
+    data = data.reshape(-1, dim + 1)
+    data = np.ascontiguousarray(data[:, 1:])
+    ndata, dim = data.shape
+    return data, ndata, dim

process/requirements.txt

@@ -0,0 +1,2 @@
+numpy
+gensim

sift1m/loading.py

@@ -1,14 +0,0 @@
-import numpy as np
-
-def fvecs_read(fname):
-   data = np.fromfile(fname, dtype=np.float32)
-   dim = data[0].view(np.int32)
-   data = data.reshape(-1, dim + 1)
-   data = np.ascontiguousarray(data[:, 1:])
-   ndata, dim = data.shape 
-   return data, ndata, dim
-
-
-# f = 'sift_base.fvecs'
-# sift = fvecs_read(f)
-# print(f"{sift.shape}")