Fixed a bug when processing data without socre columns.

blackbox/getsampledata.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+wget http://research.microsoft.com/en-us/um/beijing/projects/letor/LETOR4.0/Data/MQ2008.rar
+#apt-get install unrar-free
+unrar MQ2008.rar
+rm MQ2008.rar
+mv -f MQ2008/Fold1/*.txt .

blackbox/listnet.py

@@ -65,10 +65,9 @@ def prepare_data(self,filename=None,T=500, dim = 30, train = True, noscore = Tru
                     self.test_X = []
                     for t in data.session.unique():
                         dt = data[data.session == t]
-                        if dt.score.sum() == 0:
-                            continue
                         xt = dt.loc[:,np.in1d(dt.columns,["session"],invert=True)]
-                        self.test_X.append(xt.values.astype(np.float32))                
+                        self.test_X.append(xt.values.astype(np.float32))
+                    self.T = len(self.X)
                 else:
                     self.test_X = []
                     self.test_Y = []
@@ -171,6 +170,7 @@ def predict(self,test_X):
     def test(self,filename,noscore = True):
         testres = 0
         self.prepare_data(filename = filename, train = False, noscore = noscore)
+
         if noscore:
             pred = pd.DataFrame()
             for j in range(self.test_T):

blackbox/sample.ipynb

@@ -0,0 +1,263 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import numpy as np\n",
+    "import sklearn.externals.six\n",
+    "from sklearn.externals.six.moves import range"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "def iter_lines(lines, has_targets=True, one_indexed=True, missing=0.0):\n",
+    "    \"\"\"Transforms an iterator of lines to an iterator of LETOR rows.\n",
+    "    Each row is represented by a (x, y, qid, comment) tuple.\n",
+    "    Parameters\n",
+    "    ----------\n",
+    "    lines : iterable of lines\n",
+    "        Lines to parse.\n",
+    "    has_targets : bool, optional\n",
+    "        Whether the file contains targets. If True, will expect the first token\n",
+    "        of every line to be a real representing the sample's target (i.e.\n",
+    "        score). If False, will use -1 as a placeholder for all targets.\n",
+    "    one_indexed : bool, optional\n",
+    "        Whether feature ids are one-indexed. If True, will subtract 1 from each\n",
+    "        feature id.\n",
+    "    missing : float, optional\n",
+    "        Placeholder to use if a feature value is not provided for a sample.\n",
+    "    Yields\n",
+    "    ------\n",
+    "    x : array of floats\n",
+    "        Feature vector of the sample.\n",
+    "    y : float\n",
+    "        Target value (score) of the sample, or -1 if no target was parsed.\n",
+    "    qid : object\n",
+    "        Query id of the sample. This is currently guaranteed to be a string.\n",
+    "    comment : str\n",
+    "        Comment accompanying the sample.\n",
+    "    \"\"\"\n",
+    "    for line in lines:\n",
+    "        data, _, comment = line.rstrip().partition('#')\n",
+    "        toks = data.split()\n",
+    "\n",
+    "        num_features = 0\n",
+    "        x = np.repeat(missing, 8)\n",
+    "        y = -1.0\n",
+    "        if has_targets:\n",
+    "            y = float(toks[0])\n",
+    "            toks = toks[1:]\n",
+    "\n",
+    "        qid = _parse_qid_tok(toks[0])\n",
+    "\n",
+    "        for tok in toks[1:]:\n",
+    "            fid, _, val = tok.partition(':')\n",
+    "            fid = int(fid)\n",
+    "            val = float(val)\n",
+    "            if one_indexed:\n",
+    "                fid -= 1\n",
+    "            assert fid >= 0\n",
+    "            while len(x) <= fid:\n",
+    "                orig = len(x)\n",
+    "                x.resize(len(x) * 2)\n",
+    "                x[orig:orig * 2] = missing\n",
+    "\n",
+    "            x[fid] = val\n",
+    "            num_features = max(fid + 1, num_features)\n",
+    "\n",
+    "        assert num_features > 0\n",
+    "        x.resize(num_features)\n",
+    "\n",
+    "        yield (x, y, qid, comment)\n",
+    "\n",
+    "\n",
+    "def read_dataset(source, has_targets=True, one_indexed=True, missing=0.0):\n",
+    "    \"\"\"Parses a LETOR dataset from `source`.\n",
+    "    Parameters\n",
+    "    ----------\n",
+    "    source : string or iterable of lines\n",
+    "        String, file, or other file-like object to parse.\n",
+    "    has_targets : bool, optional\n",
+    "        See `iter_lines`.\n",
+    "    one_indexed : bool, optional\n",
+    "        See `iter_lines`.\n",
+    "    missing : float, optional\n",
+    "        See `iter_lines`.\n",
+    "    Returns\n",
+    "    -------\n",
+    "    X : array of arrays of floats\n",
+    "        Feature matrix (see `iter_lines`).\n",
+    "    y : array of floats\n",
+    "        Target vector (see `iter_lines`).\n",
+    "    qids : array of objects\n",
+    "        Query id vector (see `iter_lines`).\n",
+    "    comments : array of strs\n",
+    "        Comment vector (see `iter_lines`).\n",
+    "    \"\"\"\n",
+    "    if isinstance(source, sklearn.externals.six.string_types):\n",
+    "        source = source.splitlines()\n",
+    "\n",
+    "    max_width = 0\n",
+    "    xs, ys, qids, comments = [], [], [], []\n",
+    "    it = iter_lines(source, has_targets=has_targets,\n",
+    "                    one_indexed=one_indexed, missing=missing)\n",
+    "    for x, y, qid, comment in it:\n",
+    "        xs.append(x)\n",
+    "        ys.append(y)\n",
+    "        qids.append(qid)\n",
+    "        comments.append(comment)\n",
+    "        max_width = max(max_width, len(x))\n",
+    "\n",
+    "    assert max_width > 0\n",
+    "    X = np.ndarray((len(xs), max_width), dtype=np.float64)\n",
+    "    X.fill(missing)\n",
+    "    for i, x in enumerate(xs):\n",
+    "        X[i, :len(x)] = x\n",
+    "    ys = np.array(ys) if has_targets else None\n",
+    "    qids = np.array(qids)\n",
+    "    comments = np.array(comments)\n",
+    "\n",
+    "    return (X, ys, qids, comments)\n",
+    "\n",
+    "\n",
+    "def _parse_qid_tok(tok):\n",
+    "    assert tok.startswith('qid:')\n",
+    "    return tok[4:]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "with open('train.txt') as trainfile, \\\n",
+    "        open('vali.txt') as valifile, \\\n",
+    "        open('test.txt') as evalfile:\n",
+    "    TX, Ty, Tqids, _ = read_dataset(trainfile)\n",
+    "    VX, Vy, Vqids, _ =read_dataset(valifile)\n",
+    "    EX, Ey, Eqids, _ = read_dataset(evalfile)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "traindata =pd.concat([pd.DataFrame(TX) ,pd.DataFrame(Ty,columns=[\"score\"]),pd.DataFrame(Tqids,columns=[\"session\"])],axis=1)\n",
+    "validata =pd.concat([pd.DataFrame( VX) ,pd.DataFrame(Vy,columns=[\"score\"]),pd.DataFrame(Vqids,columns=[\"session\"])],axis=1)\n",
+    "testdata =pd.concat([pd.DataFrame(EX)  ,pd.DataFrame(Eqids,columns=[\"session\"])],axis=1)\n",
+    "traindata.to_csv(\"datasets.csv\",index=None)\n",
+    "validata.to_csv(\"datasets_test.csv\",index=None)\n",
+    "testdata.to_csv(\"datasets_test_noscore.csv\",index=None)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "step:9,train_loss:7.46409893036\n",
+      "train_ndcg:0.812323686008\n",
+      "step:19,train_loss:6.07994031906\n",
+      "train_ndcg:0.909060459757\n",
+      "step:29,train_loss:7.78203868866\n",
+      "train_ndcg:0.944716231404\n",
+      "step:39,train_loss:6.72615432739\n",
+      "train_ndcg:0.966039979587\n",
+      "step:49,train_loss:7.4245595932\n",
+      "train_ndcg:0.977167381069\n",
+      "step:59,train_loss:9.81043434143\n",
+      "train_ndcg:0.985105804125\n",
+      "step:69,train_loss:8.38823699951\n",
+      "train_ndcg:0.989879857973\n"
+     ]
+    }
+   ],
+   "source": [
+    "!python2.7 listnet.py --train_val_filename \"datasets.csv\" --test_score_filename \"datasets_test.csv\" --test_noscore_filename \"datasets_test_noscore.csv\" --val_ratio 0.0 --max_iter 100\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "anaconda-cloud": {},
+  "kernelspec": {
+   "display_name": "Python [py35]",
+   "language": "python",
+   "name": "Python [py35]"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.5.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}