Performance work (#334)

* range encode filter range lists

* speed up data load

* add comment on scanpy read params

* update to latest scanpy/anndata

* performance improvments in data loading

* fix typo

* work around scanpy bug

* remove debugging print statements

client/__tests__/util/actionHelpers.test.js

@@ -0,0 +1,60 @@
+import { rangeEncodeIndices } from "../../src/util/actionHelpers";
+
+describe("rangeEncodeIndices", () => {
+  test("small array edge cases", () => {
+    expect(rangeEncodeIndices([])).toMatchObject([]);
+    expect(rangeEncodeIndices([1])).toMatchObject([1]);
+    expect(rangeEncodeIndices([1, 99])).toMatchObject([1, 99]);
+    expect(rangeEncodeIndices([99, 1])).toMatchObject([1, 99]);
+    expect(rangeEncodeIndices([1, 100, 4])).toMatchObject([1, 4, 100]);
+  });
+
+  test("sorted flag", () => {
+    expect(rangeEncodeIndices([1, 9, 432], 10, true)).toMatchObject([
+      1,
+      9,
+      432
+    ]);
+    expect(rangeEncodeIndices([1, 9, 432], 10, false)).toMatchObject([
+      1,
+      9,
+      432
+    ]);
+    expect(rangeEncodeIndices([0, 1, 2, 3, 9, 10, 432], 2, true)).toMatchObject(
+      [[0, 3], [9, 10], 432]
+    );
+    expect(
+      rangeEncodeIndices([0, 1, 2, 3, 9, 10, 432], 2, false)
+    ).toMatchObject([[0, 3], [9, 10], 432]);
+  });
+
+  test("begin or end edge cases", () => {
+    expect(
+      rangeEncodeIndices([3, 4, 5, 6, 7, 10, 11, 12, 99], 3, false)
+    ).toMatchObject([[3, 7], [10, 12], 99]);
+    expect(
+      rangeEncodeIndices([3, 4, 5, 6, 7, 10, 11, 12], 3, false)
+    ).toMatchObject([[3, 7], [10, 12]]);
+    expect(
+      rangeEncodeIndices([0, 3, 4, 5, 6, 7, 10, 11, 12], 3, false)
+    ).toMatchObject([0, [3, 7], [10, 12]]);
+    expect(
+      rangeEncodeIndices([0, 3, 4, 5, 6, 7, 10, 11, 12, 99], 3, false)
+    ).toMatchObject([0, [3, 7], [10, 12], 99]);
+  });
+
+  test("minRangeLength", () => {
+    expect(
+      rangeEncodeIndices([3, 4, 5, 6, 7, 10, 11, 12, 99], 4, false)
+    ).toMatchObject([[3, 7], 10, 11, 12, 99]);
+    expect(
+      rangeEncodeIndices([3, 4, 5, 6, 7, 10, 11, 12], 4, false)
+    ).toMatchObject([[3, 7], 10, 11, 12]);
+    expect(
+      rangeEncodeIndices([0, 3, 4, 5, 6, 7, 10, 11, 12], 4, false)
+    ).toMatchObject([0, [3, 7], 10, 11, 12]);
+    expect(
+      rangeEncodeIndices([0, 3, 4, 5, 6, 7, 10, 11, 12, 99], 4, false)
+    ).toMatchObject([0, [3, 7], 10, 11, 12, 99]);
+  });
+});

client/src/actions/index.js

@@ -2,8 +2,18 @@
 import _ from "lodash";
 import * as globals from "../globals";
 import { Universe, kvCache } from "../util/stateManager";
-import { catchErrorsWrap, doJsonRequest } from "../util/actionHelpers";
+import {
+  catchErrorsWrap,
+  doJsonRequest,
+  rangeEncodeIndices
+} from "../util/actionHelpers";
 
+/*
+Bootstrap application with the initial data loading.
+  * /config - application configuration
+  * /schema - schema of dataframe
+  * /annotations/obs - all metadata annotation
+*/
 const doInitialDataLoad = () =>
   catchErrorsWrap(async dispatch => {
     dispatch({ type: "initial data load start" });
@@ -235,8 +245,12 @@ const requestDifferentialExpression = (set1, set2, num_genes = 10) => async (
     */
     const state = getState();
     const { universe } = state.controls;
-    const set1ByIndex = _.map(set1, s => universe.obsNameToIndexMap[s]);
+    const set1ByIndex = rangeEncodeIndices(
-    const set2ByIndex = _.map(set2, s => universe.obsNameToIndexMap[s]);
+      _.map(set1, s => universe.obsNameToIndexMap[s])
+    );
+    const set2ByIndex = rangeEncodeIndices(
+      _.map(set2, s => universe.obsNameToIndexMap[s])
+    );
     const diffExpFetch = await fetch(
       `${globals.API.prefix}${globals.API.version}diffexp/obs`,
       {

client/src/util/actionHelpers.js

@@ -1,3 +1,5 @@
+import _ from "lodash";
+
 /*
 Catch unexpected errors and make sure we don't lose them!
 */
@@ -11,10 +13,7 @@ export function catchErrorsWrap(fn) {
 }
 
 /*
-Bootstrap application with the initial data loading.
+Wrapper to perform an async fetch and JSON decode response.
-  * /config - application configuration
-  * /schema - schema of dataframe
-  * /annotations/obs - all metadata annotation
 */
 export const doJsonRequest = async url => {
   const res = await fetch(url, {
@@ -26,3 +25,57 @@ export const doJsonRequest = async url => {
   });
   return res.json();
 };
+
+/*
+This function "packs" filter index lists into the more efficient
+"range" form specified in the REST 0.2 spec.
+
+Specifically, it turns an array of indices [0, 1, 2, 10, 11, 14, ...]
+into a form that encodes runs of consecutive numbers as [min, max].
+Array may not be sorted, but will only contain uniq values.
+
+Parameters:
+   indices - input array of numbers (index)
+   minRangeLength - hint, min range length before it is encoded into range format.
+   sorted - boolean hint indicating array is presorted, ascending order
+
+So [1, 2, 3, 4, 10, 11, 14] -> [ [1, 4], [10, 11], 14]
+*/
+export const rangeEncodeIndices = (
+  indices,
+  minRangeLength = 3,
+  sorted = false
+) => {
+  if (indices.length === 0) {
+    return indices;
+  }
+
+  if (!sorted) {
+    indices = _.sortBy(indices);
+  }
+
+  const result = new Array(indices.length);
+  let resultTail = 0;
+
+  let i = 0;
+  while (i < indices.length) {
+    const begin = indices[i];
+    let current;
+    do {
+      current = indices[i];
+      i += 1;
+    } while (i < indices.length && indices[i] === current + 1);
+
+    if (current - begin + 1 >= minRangeLength) {
+      result[resultTail] = [begin, current];
+      resultTail += 1;
+    } else {
+      for (let j = begin; j <= current; j += 1, resultTail += 1) {
+        result[resultTail] = j;
+      }
+    }
+  }
+
+  result.length = resultTail;
+  return result;
+};

server/app/scanpy_engine/scanpy_engine.py

@@ -30,7 +30,7 @@ class ScanpyEngine(CXGDriver):
         self.cell_count = self.data.shape[0]
         self.gene_count = self.data.shape[1]
         self._create_schema()
-        self.layout(None)
+        self.layout({})
 
     def _create_schema(self):
         self.schema = {
@@ -78,7 +78,11 @@ class ScanpyEngine(CXGDriver):
 
     @staticmethod
     def _load_data(data):
-        return sc.read(os.path.join(data, "data.h5ad"))
+        # See https://scanpy.readthedocs.io/en/latest/api/scanpy.api.read.html
+        # Based upon this advice, setting cache=True parameter
+        # Note: as of current scanpy/anndata release, setting backed='r' will
+        # result in an error.
+        return sc.read(os.path.join(data, "data.h5ad"), cache=True)
 
     @staticmethod
     def _top_sort(values, sort_order, top_n=None):

server/requirements.txt

@@ -1,4 +1,4 @@
-anndata==0.6.10
+anndata==0.6.11
 Flask==0.12.4
 Flask-Caching==1.4.0
 Flask-Compress==1.4.0
@@ -7,5 +7,5 @@ Flask-RESTful==0.3.6
 flask-restful-swagger-2==0.35
 numpy==1.14.5
 pandas==0.23.1
-scanpy==1.3.1
+scanpy==1.3.2
 scipy==1.1.0