X float16 support (#2406)

* float16 support

* fix type checks

* PR review comments

* add tests for custom json encoder; rename and comment for posterity

* lint

* typos

backend/common/compute/estimate_distribution.py

@@ -1,3 +1,4 @@
+from typing import Tuple
 import numba
 import concurrent.futures
 import numpy as np
@@ -6,7 +7,7 @@ from backend.common.constants import XApproximateDistribution
 
 
 @numba.njit(error_model="numpy", nogil=True)
-def min_max(arr: np.ndarray):
+def min_max_fast(arr: np.ndarray) -> Tuple[float, float]:
     """Return (min, max) values for the ndarray."""
 
     # initialize to first finite value in array.  Normally,
@@ -47,6 +48,24 @@ def min_max(arr: np.ndarray):
     return min_val, max_val
 
 
+def min_max_numpy(arr: np.ndarray) -> Tuple[float, float]:
+    return arr.min(), arr.max()
+
+
+def numba_has_support_for_scalar_type(arr: np.ndarray) -> bool:
+    """Numba does not support half-floats, 128 bit floats, ints > 64 bit or non-scalars."""
+    if arr.dtype == np.float32 or arr.dtype == np.float64:
+        return True
+
+    if np.issubdtype(arr.dtype, np.integer) and arr.dtype <= np.int64:
+        return True
+
+    if arr.dtype == np.bool_:
+        return True
+
+    return False
+
+
 def estimate_approximate_distribution(X) -> XApproximateDistribution:
     """
     Estimate the distribution (normal, count) of the X matrix.
@@ -72,6 +91,8 @@ def estimate_approximate_distribution(X) -> XApproximateDistribution:
     else:
         raise TypeError(f"Unsupported matrix format: {str(type(X))}")
 
+    min_max = min_max_fast if numba_has_support_for_scalar_type(Xdata) else min_max_numpy
+
     CHUNKSIZE = 1 << 24
     if Xdata.size > CHUNKSIZE:
         min_val = max_val = Xdata[0]

backend/common/utils/utils.py

@@ -65,7 +65,13 @@ def path_join(base, *urls):
     return btpl._replace(path=path).geturl()
 
 
-class Float32JSONEncoder(json.JSONEncoder):
+class StrictJSONEncoder(json.JSONEncoder):
+    """
+    Custom JSON encoder set-up performing two tasks:
+    1. Strict JSON conformance with non-finite floats (NaN, +/-Inf) via allow_nan=False
+    2. Convert various Numpy types into python types so the encoder will correctly encode.
+    """
+
     def __init__(self, *args, **kwargs):
         """
         NaN/Infinities are illegal in standard JSON.  Python extends JSON with
@@ -78,9 +84,11 @@ class Float32JSONEncoder(json.JSONEncoder):
         super().__init__(*args, **kwargs)
 
     def default(self, obj):
-        if isinstance(obj, np.float32):
+        """This helps us convert types not supported by the native JSON encoder into
+        standard python types, eg, np.int64."""
+        if isinstance(obj, np.floating):
             return float(obj)
-        elif isinstance(obj, np.integer):
+        if isinstance(obj, np.integer):
             return int(obj)
         return json.JSONEncoder.default(self, obj)
 
@@ -89,8 +97,8 @@ def custom_format_warning(msg, *args, **kwargs):
     return f"[cellxgene] Warning: {msg} \n"
 
 
-def jsonify_numpy(data):
-    return json.dumps(data, cls=Float32JSONEncoder, allow_nan=False)
+def jsonify_strict(data):
+    return json.dumps(data, cls=StrictJSONEncoder, allow_nan=False)
 
 
 def import_plugins(plugin_module):

backend/czi_hosted/app/app.py

@@ -24,7 +24,7 @@ import backend.czi_hosted.common.rest as common_rest
 from backend.common.utils.data_locator import DataLocator
 from backend.common.errors import DatasetAccessError, RequestException
 from backend.czi_hosted.common.health import health_check
-from backend.common.utils.utils import path_join, Float32JSONEncoder
+from backend.common.utils.utils import path_join, StrictJSONEncoder
 from backend.czi_hosted.data_common.matrix_loader import MatrixDataLoader
 
 webbp = Blueprint("webapp", "backend.czi_hosted.common.web", template_folder="templates")
@@ -396,7 +396,7 @@ class Server:
         self.app = Flask(__name__, static_folder=None)
         handle_api_base_url(self.app, app_config)
         self._before_adding_routes(self.app, app_config)
-        self.app.json_encoder = Float32JSONEncoder
+        self.app.json_encoder = StrictJSONEncoder
         server_config = app_config.server_config
         if server_config.app__server_timing_headers:
             ServerTiming(self.app, force_debug=True)

backend/czi_hosted/data_common/data_adaptor.py

@@ -15,7 +15,7 @@ from backend.common.errors import (
     UnsupportedSummaryMethod,
     DatasetAccessError,
 )
-from backend.common.utils.utils import jsonify_numpy
+from backend.common.utils.utils import jsonify_strict
 from backend.common.fbs.matrix import encode_matrix_fbs
 
 
@@ -336,7 +336,7 @@ class DataAdaptor(metaclass=ABCMeta):
         )
 
         try:
-            return jsonify_numpy(result)
+            return jsonify_strict(result)
         except ValueError:
             raise JSONEncodingValueError("Error encoding differential expression to JSON")
 

backend/server/app/app.py

@@ -17,7 +17,7 @@ from flask_restful import Api, Resource
 import backend.server.common.rest as common_rest
 from backend.common.errors import DatasetAccessError, RequestException
 from backend.server.common.health import health_check
-from backend.common.utils.utils import Float32JSONEncoder
+from backend.common.utils.utils import StrictJSONEncoder
 
 webbp = Blueprint("webapp", "backend.server.common.web", template_folder="templates")
 
@@ -257,7 +257,7 @@ class Server:
     def __init__(self, app_config):
         self.app = Flask(__name__, static_folder=None)
         self._before_adding_routes(self.app, app_config)
-        self.app.json_encoder = Float32JSONEncoder
+        self.app.json_encoder = StrictJSONEncoder
         server_config = app_config.server_config
 
         # enable session data

backend/server/data_anndata/anndata_adaptor.py

@@ -232,7 +232,7 @@ class AnndataAdaptor(DataAdaptor):
                 "Anndata data matrix is sparse, but not a CSC (columnar) matrix.  "
                 "Performance may be improved by using CSC."
             )
-        if self.data.X.dtype != "float32":
+        if self.data.X.dtype > np.dtype(np.float32):
             warnings.warn(
                 f"Anndata data matrix is in {self.data.X.dtype} format not float32. " f"Precision may be truncated."
             )

backend/server/data_common/data_adaptor.py

@@ -8,7 +8,7 @@ from server_timing import Timing as ServerTiming
 from backend.server.common.config.app_config import AppConfig
 from backend.common.constants import Axis, XApproximateDistribution
 from backend.common.errors import FilterError, JSONEncodingValueError, ExceedsLimitError, UnsupportedSummaryMethod
-from backend.common.utils.utils import jsonify_numpy
+from backend.common.utils.utils import jsonify_strict
 from backend.common.fbs.matrix import encode_matrix_fbs
 from backend.common.genesets import validate_gene_sets
 
@@ -331,7 +331,7 @@ class DataAdaptor(metaclass=ABCMeta):
         )
 
         try:
-            return jsonify_numpy(result)
+            return jsonify_strict(result)
         except ValueError:
             raise JSONEncodingValueError("Error encoding differential expression to JSON")
 

backend/test/test_common/unit/utils/test_jsonify_strict.py

@@ -0,0 +1,58 @@
+import unittest
+import numpy as np
+
+from backend.common.utils.utils import (
+    jsonify_strict,
+)
+
+
+class TestJsonifyStrict(unittest.TestCase):
+    def test_jsonify_numpy_general_cases(self):
+        self.assertEqual(jsonify_strict({}), "{}")
+        self.assertEqual(jsonify_strict({"a": [], "b": "hello", "c": True}), '{"a": [], "b": "hello", "c": true}')
+
+    def test_jsonify_numpy_float_edges(self):
+        with self.assertRaises(ValueError):
+            jsonify_strict({"nan": [np.nan]})
+
+        with self.assertRaises(ValueError):
+            jsonify_strict({"pinf": [np.PINF]})
+
+        with self.assertRaises(ValueError):
+            jsonify_strict({"ninf": [np.NINF]})
+
+    def test_jsonify_numpy_ndarray(self):
+        values = {
+            "integer": [
+                np.int8(0),
+                np.int16(1),
+                np.int32(2),
+                np.int64(3),
+                np.uint8(4),
+                np.uint16(5),
+                np.uint32(6),
+                np.uint64(7),
+            ],
+            "floating": [
+                np.float16(100.0),
+                np.float32(101.0),
+                np.float64(102.0),
+            ],
+        }
+        # these just confirm our test assumptions
+        self.assertTrue(isinstance(values["floating"][0], np.float16))
+        self.assertTrue(isinstance(values["floating"][1], np.float32))
+        self.assertTrue(isinstance(values["floating"][2], np.float64))
+        self.assertTrue(isinstance(values["integer"][0], np.int8))
+        self.assertTrue(isinstance(values["integer"][1], np.int16))
+        self.assertTrue(isinstance(values["integer"][2], np.int32))
+        self.assertTrue(isinstance(values["integer"][3], np.int64))
+        self.assertTrue(isinstance(values["integer"][4], np.uint8))
+        self.assertTrue(isinstance(values["integer"][5], np.uint16))
+        self.assertTrue(isinstance(values["integer"][6], np.uint32))
+        self.assertTrue(isinstance(values["integer"][7], np.uint64))
+        # the actual test!
+        self.assertEqual(
+            jsonify_strict(values),
+            '{"floating": [100.0, 101.0, 102.0], "integer": [0, 1, 2, 3, 4, 5, 6, 7]}',
+        )