Added implementation of nunique function (#29)

* Added implementation of nunique function

* Added test for handling strings nulls (" "), differentiating behavior between Python and kdb+

* Suggested changes. Error with mixed lists and tests for this case.

* QError for mixed lists (suggested by Kx)

* minor: rename filternan (suggested)

docs/user-guide/advanced/Pandas_API.ipynb

@@ -2535,7 +2535,48 @@
   },
   {
    "cell_type": "markdown",
-   "id": "499025cb",
+   "source": [
+    "### Table.nunique()\n",
+    "```\n",
+    "Table.nunique(axis=0, skipna=True, numeric_only=False, min_count=0)\n",
+    "```\n",
+    "\n",
+    "Returns the number of unique elements across the given axis.\n",
+    "\n",
+    "**Parameters:**\n",
+    "\n",
+    "| Name         | Type | Description                                                                         | Default |\n",
+    "| :----------: | :--: |:------------------------------------------------------------------------------------| :-----: |\n",
+    "| axis         | int  | The axis to calculate the number of unique elements across 0 is columns, 1 is rows. | 0       |\n",
+    "| dropna       | bool | Don’t include NaN in the counts.                                                    | True    |\n",
+    "\n",
+    "**Returns:**\n",
+    "\n",
+    " | Type               | Description                                                          |\n",
+    " | :----------------: | :------------------------------------------------------------------- |\n",
+    " | Dictionary         | A dictionary where the key represent the column name / row number and the values are the result of calling `nunique` on that column / row. |"
+   ],
+   "metadata": {
+    "collapsed": false
+   },
+   "id": "5bc5e813e9673a84"
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "outputs": [],
+   "source": [
+    "tab.nunique()"
+   ],
+   "metadata": {
+    "collapsed": false
+   },
+   "id": "f5592b19b69ad46d"
+  },
+  {
+   "cell_type": "markdown",
+   "id": "655c3ad2",
+
    "metadata": {},
    "source": [
     "## Setting Indexes"

src/pykx/pandas_api/pandas_meta.py

@@ -261,6 +261,14 @@ def sum(self, axis=0, skipna=True, numeric_only=False, min_count=0):
             min_count
         ), cols)
 
+    @convert_result
+    def nunique(self, axis=0, dropna=True):
+        res, cols = preparse_computations(self, axis, skipna=False)
+        filternan = q('{$[all[10h=type each x]|11h = type x;x;'
+                      'x where not null x]}each')
+        res = filternan(res) if dropna else res
+        return (q("('[count;distinct]')", res), cols)
+
     def agg(self, func, axis=0, *args, **kwargs): # noqa: C901
         if 'KeyedTable' in str(type(self)):
             raise NotImplementedError("'agg' method not presently supported for KeyedTable")

tests/test_pandas_api.py

@@ -2038,3 +2038,37 @@ def test_keyed_loc_fixes(q):
         mkt[['k1', 'y']]
     with pytest.raises(KeyError):
         mkt['k1']
+
+
+def test_nunique(kx, q):
+    tab = kx.q('([]a:4 0n 7 6;b:4 0n 0n 7;c:``foo`foo`)')
+    df = tab.pd()
+    p_m = df.nunique()
+    q_m = tab.nunique()
+    for c in q.key(q_m).py():
+        assert p_m[c] == q_m[c].py()
+    p_m = df.nunique(dropna=False)
+    q_m = tab.nunique(dropna=False)
+    for c in q.key(q_m).py():
+        assert p_m[c] == q_m[c].py()
+
+    df = pd.DataFrame(
+        {
+            'a': [1, 2, 2, 4],
+            'b': [1, 2, 6, 7],
+            'c': [7, 8, 9, 10],
+        }
+    )
+    tab = kx.toq(df)
+    p_m = df.nunique()
+    q_m = tab.nunique()
+    for c in q.key(q_m).py():
+        assert p_m[c] == q_m[c].py()
+    p_m = df.nunique(axis=1)
+    q_m = tab.nunique(axis=1)
+    for c in range(len(tab)):
+        assert p_m[c] == q_m[c].py()
+
+    tab = kx.q('([]a:("";" ";"";"foo"))')
+    with pytest.raises(kx.QError):
+        raise tab.nunique()