Merge branch 'master' of github.com:adamhrv/megapixels_dev

3 files changed, 91 insertions, 0 deletions

diff --git a/megapixels/commands/faiss/build_db.py b/megapixels/commands/faiss/build_db.py
new file mode 100644
index 00000000..0f979e41
--- /dev/null
+++ b/megapixels/commands/faiss/build_db.py
@@ -0,0 +1,15 @@
+"""
+Load all the CSV files into MySQL
+"""
+
+import click
+
+from app.models.sql_factory import load_sql_datasets
+
+@click.command()
+@click.pass_context
+def cli(ctx):
+  """import the various CSVs into MySQL
+  """
+  print('Loading CSV datasets into SQL...')
+  load_sql_datasets(replace=True)
diff --git a/megapixels/commands/faiss/build_faiss.py b/megapixels/commands/faiss/build_faiss.py
new file mode 100644
index 00000000..96d3f99e
--- /dev/null
+++ b/megapixels/commands/faiss/build_faiss.py
@@ -0,0 +1,58 @@
+"""
+Index all of the FAISS datasets
+"""
+
+import os
+import glob
+import click
+import faiss
+import time
+import numpy as np
+
+from app.utils.file_utils import load_recipe, load_csv_safe
+from app.settings import app_cfg as cfg
+
+engine = create_engine('sqlite:///:memory:')
+
+class DefaultRecipe:
+  def __init__(self):
+    self.dim = 128
+    self.factory_type = 'Flat'
+
+@click.command()
+@click.pass_context
+def cli(ctx):
+  """build the FAISS index.
+    - looks for all datasets in faiss/metadata/
+    - uses the recipe above by default
+    - however you can override this by adding a new recipe in faiss/recipes/{name}.json
+  """
+  datasets = []
+  for fn in glob.iglob(os.path.join(cfg.DIR_FAISS_METADATA, "*")):
+    name = os.path.basename(fn)
+    recipe_fn = os.path.join(cfg.DIR_FAISS_RECIPES, name + ".json")
+    if os.path.exists(recipe_fn):
+      build_faiss(name, load_recipe(recipe_fn))
+    else:
+      build_faiss(name, DefaultRecipe())
+
+def build_faiss(name, recipe):
+  vec_fn = os.path.join(cfg.DIR_FAISS_METADATA, name, "vecs.csv")
+  index_fn = os.path.join(cfg.DIR_FAISS_INDEXES, name + ".index")
+
+  index = faiss.index_factory(recipe.dim, recipe.factory_type)
+
+  keys, rows = load_csv_safe(vec_fn)
+  feats = np.array([ list(map(float, row[3].split(","))) for row in rows ]).astype('float32')
+  n, d = feats.shape
+
+  print("{}: training {} x {} dim vectors".format(name, n, d))
+  print(recipe.factory_type)
+
+  add_start = time.time()
+  index.add(feats)
+  add_end = time.time()
+  add_time = add_end - add_start
+  print("{}: add time: {:.1f}s".format(name, add_time))
+
+  faiss.write_index(index, index_fn)
diff --git a/megapixels/commands/faiss/sync_metadata.py b/megapixels/commands/faiss/sync_metadata.py
new file mode 100644
index 00000000..b01211b4
--- /dev/null
+++ b/megapixels/commands/faiss/sync_metadata.py
@@ -0,0 +1,18 @@
+"""
+Sync the FAISS metadata
+"""
+
+import subprocess
+import click
+
+from app.settings import app_cfg as cfg
+
+@click.command()
+@click.pass_context
+def cli(ctx):
+  """synchronize metadata files from s3"""
+  sts = subprocess.call([
+    "s3cmd", "sync",
+    "s3://megapixels/v1/metadata/",
+    cfg.DIR_FAISS_METADATA + '/',
+  ])