Beginner question: How to predict from Matrix Factorization model

1

Hi all,

I’m new to MxNet and following the tutorial about movie rating predictions based on Matrix Factorization. I now have a trained model, however, unfortunately the tutorial stops there. It’s not clear to me how to use this model to make predictions. From what I understand I can give it a user id and a movie title id, and it should give me the predicted rating. How do I do that?

Please note that I’m using mxnet.module.module.Module instead of FeedForward. I tried using model.forward and model.predict, but it’s not clear to me what to pass in.

Appreciate any help

Thanks

2

Module.predict will give you the prediction result. You need to pass in a data iterator for the data you want to predict. What did you try and did it not work?

3 
data = mx.nd.array([1,100])
labelToPredict = mx.nd.array(['score'])
dataiter = mx.io.NDArrayIter(data, labelToPredict);

model.predict(eval_data=dataiter, num_batch=1)

I’m getting “TypeError: source_array must be array like object” with the above. But it’s probably wrong anyway. What I want is give me the predicted score of the user 1 for movie 100. How do I properly construct a data array for this?

4

What you’d want to do is to feed in a DataIter object that contains "item" and "user" data in a dictionary. In the notebook that you’re following, if you change the code so that the train() function returns the model, then you can call predict for user 1 and movie 100 with the following code:

data = {'user':mx.nd.array([1]), 'item':mx.nd.array([100])}
data_iter = mx.io.NDArrayIter(data, batch_size=1)
score = model.predict(data_iter, num_batch=1)

You can easily modify the code to do batch predictions as well:

data = {'user':np.random.randint(0, max_user, size=(1024,)), 'item':np.random.randint(0, max_item, size=(1024,))}
data_iter = mx.io.NDArrayIter(data, batch_size=128)
model.predict(data_iter, num_batch=1024//128)

I must also mention that this tutorial is using the deprecated FeedForward API, which is replaced with the module API. You may want to use the tutorial found here instead. Also since you seem to be just starting to learn MXNet, I highly recommend focusing on the Gluon API, which is an imperative API and is much easier to learn and debug.

5

Thanks a lot for your reply. I feel like I’m almost there. When I run the code you suggested, I get the following error:

AssertionError: Shape of unspecified array arg:score changed. This can cause the new executor to not share parameters with the old one. Please check for error in network.If this is intended, set partial_shaping=True to suppress this warning.

I assume I have to tell it the shape of the label score (a single value). But I can’t figure out how to do that. Thoughts?

6

I’m not really sure how you’re getting that error because I don’t see it. In the following example code, I just changed the Train() code to return the model at the end of the function and I added the following predict function:

def train(network, batch_size, num_epoch, learning_rate):
    # Exisitng code for train
    .
    .
    .
    return model

def predict(net, users, items, batch_size):
    assert users.shape == items.shape and users.ndim == 1
    data = {'user':users, 'item':items}
    data_iter = mx.io.NDArrayIter(data, batch_size=batch_size)
    return net.predict(data_iter)

Then I modify the call to train() so that the model is saved and then passed to predict():

model = train(get_one_layer_mlp(64, 64), batch_size=64, num_epoch=10, learning_rate=.05)
print(predict(
    model,
    users=np.random.randint(0, max_user, size=(16,)),
    items=np.random.randint(0, max_item, size=(16,)),
    batch_size=4))

If this doesn’t fix your problem, feel free to paste your code here so I can see how you’re getting that error.

7

Ok, here’s the notebook:

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Matrix Factorization\n",
    "\n",
    "In a recommendation system, there is a group of users and a set of items. Given that each users have rated some items in the system, we would like to predict how the users would rate the items that they have not yet rated, such that we can make recommendations to the users.\n",
    "\n",
    "Matrix factorization is one of the mainly used algorithm in recommendation systems. It can be used to discover latent features underlying the interactions between two different kinds of entities.\n",
    "\n",
    "Assume we assign a k-dimensional vector to each user and a k-dimensional vector to each item such that the dot product of these two vectors gives the user's rating of that item. We can learn the user and item vectors directly, which is essentially performing SVD on the user-item matrix. We can also try to learn the latent features using multi-layer neural networks. \n",
    "\n",
    "In this tutorial, we will work though the steps to implement these ideas in MXNet."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prepare Data\n",
    "\n",
    "We use the [MovieLens](http://grouplens.org/datasets/movielens/) data here, but it can apply to other datasets as well. Each row of this dataset contains a tuple of user id, movie id, rating, and time stamp, we will only use the first three items. We first define the a batch which contains n tuples. It also provides name and shape information to MXNet about the data and label. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Batch(object):\n",
    "    def __init__(self, data_names, data, label_names, label):\n",
    "        self.data = data\n",
    "        self.label = label\n",
    "        self.data_names = data_names\n",
    "        self.label_names = label_names\n",
    "        \n",
    "    @property\n",
    "    def provide_data(self):\n",
    "        return [(n, x.shape) for n, x in zip(self.data_names, self.data)]\n",
    "    \n",
    "    @property\n",
    "    def provide_label(self):\n",
    "        return [(n, x.shape) for n, x in zip(self.label_names, self.label)]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then we define a data iterator, which returns a batch of tuples each time. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import mxnet as mx\n",
    "import random\n",
    "\n",
    "class Batch(object):\n",
    "    def __init__(self, data_names, data, label_names, label):\n",
    "        self.data = data\n",
    "        self.label = label\n",
    "        self.data_names = data_names\n",
    "        self.label_names = label_names\n",
    "\n",
    "    @property\n",
    "    def provide_data(self):\n",
    "        return [(n, x.shape) for n, x in zip(self.data_names, self.data)]\n",
    "\n",
    "    @property\n",
    "    def provide_label(self):\n",
    "        return [(n, x.shape) for n, x in zip(self.label_names, self.label)]\n",
    "\n",
    "class DataIter(mx.io.DataIter):\n",
    "    def __init__(self, fname, batch_size):\n",
    "        super(DataIter, self).__init__()\n",
    "        self.batch_size = batch_size\n",
    "        self.data = []\n",
    "        for line in file(fname):\n",
    "            tks = line.strip().split('\\t')\n",
    "            if len(tks) != 4:\n",
    "                continue\n",
    "            self.data.append((int(tks[0]), int(tks[1]), float(tks[2])))\n",
    "        self.provide_data = [('user', (batch_size, )), ('item', (batch_size, ))]\n",
    "        self.provide_label = [('score', (self.batch_size, ))]\n",
    "\n",
    "    def __iter__(self):\n",
    "        for k in range(len(self.data) / self.batch_size):\n",
    "            users = []\n",
    "            items = []\n",
    "            scores = []\n",
    "            for i in range(self.batch_size):\n",
    "                j = k * self.batch_size + i\n",
    "                user, item, score = self.data[j]\n",
    "                users.append(user)\n",
    "                items.append(item)\n",
    "                scores.append(score)\n",
    "\n",
    "            data_all = [mx.nd.array(users), mx.nd.array(items)]\n",
    "            label_all = [mx.nd.array(scores)]\n",
    "            data_names = ['user', 'item']\n",
    "            label_names = ['score']\n",
    "\n",
    "            data_batch = Batch(data_names, data_all, label_names, label_all)\n",
    "            yield data_batch\n",
    "\n",
    "    def reset(self):\n",
    "        random.shuffle(self.data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we download the data and provide a function to obtain the data iterator:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import urllib\n",
    "import zipfile\n",
    "if not os.path.exists('ml-100k.zip'):\n",
    "    urllib.urlretrieve('http://files.grouplens.org/datasets/movielens/ml-100k.zip', 'ml-100k.zip')\n",
    "with zipfile.ZipFile(\"ml-100k.zip\",\"r\") as f:\n",
    "    f.extractall(\"./\")\n",
    "def get_data(batch_size):\n",
    "    return (DataIter('./ml-100k/u1.base', batch_size), DataIter('./ml-100k/u1.test', batch_size))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally we calculate the numbers of users and items for later use."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(944, 1683)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def max_id(fname):\n",
    "    mu = 0\n",
    "    mi = 0\n",
    "    for line in file(fname):\n",
    "        tks = line.strip().split('\\t')\n",
    "        if len(tks) != 4:\n",
    "            continue\n",
    "        mu = max(mu, int(tks[0]))\n",
    "        mi = max(mi, int(tks[1]))\n",
    "    return mu + 1, mi + 1\n",
    "max_user, max_item = max_id('./ml-100k/u.data')\n",
    "(max_user, max_item)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Optimization\n",
    "\n",
    "We first implement the RMSE (root-mean-square error) measurement, which is commonly used by matrix factorization. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "def RMSE(label, pred):\n",
    "    ret = 0.0\n",
    "    n = 0.0\n",
    "    pred = pred.flatten()\n",
    "    for i in range(len(label)):\n",
    "        ret += (label[i] - pred[i]) * (label[i] - pred[i])\n",
    "        n += 1.0\n",
    "    return math.sqrt(ret / n)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then we define a general training module, which is borrowed from the image classification application. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(network, batch_size, num_epoch, learning_rate):\n",
    "    batch_size = batch_size\n",
    "    train, test = get_data(batch_size)\n",
    "    model = mx.mod.Module(symbol = network, \n",
    "                          data_names=[x[0] for x in train.provide_data],\n",
    "                          label_names=[y[0] for y in train.provide_label],\n",
    "                          context=[mx.cpu()])\n",
    "\n",
    "    import logging\n",
    "    head = '%(asctime)-15s %(message)s'\n",
    "    logging.basicConfig(level=logging.DEBUG)\n",
    "\n",
    "    model.fit(train_data = train, \n",
    "              eval_data = test,\n",
    "              num_epoch=num_epoch,\n",
    "              optimizer='sgd',\n",
    "              optimizer_params={'learning_rate':learning_rate, 'momentum':0.9, 'wd':0.0001},\n",
    "              eval_metric = RMSE,\n",
    "              batch_end_callback=mx.callback.Speedometer(batch_size, 20000/batch_size),)\n",
    "    \n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict(net, users, items, batch_size):\n",
    "    assert users.shape == items.shape and users.ndim == 1\n",
    "    data = {'user':users, 'item':items}\n",
    "    data_iter = mx.io.NDArrayIter(data, batch_size=batch_size)\n",
    "    return net.predict(data_iter)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Networks\n",
    "\n",
    "Now we try various networks. We first learn the latent vectors directly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Epoch[0] Batch [312]\tSpeed: 67563.78 samples/sec\tRMSE=3.721034\n",
      "INFO:root:Epoch[0] Batch [624]\tSpeed: 64852.64 samples/sec\tRMSE=3.682020\n",
      "INFO:root:Epoch[0] Batch [936]\tSpeed: 62305.78 samples/sec\tRMSE=3.625233\n",
      "INFO:root:Epoch[0] Batch [1248]\tSpeed: 68365.30 samples/sec\tRMSE=3.690150\n",
      "INFO:root:Epoch[0] Train-RMSE=3.411721\n",
      "INFO:root:Epoch[0] Time cost=1.227\n",
      "INFO:root:Epoch[0] Validation-RMSE=3.717236\n",
      "INFO:root:Epoch[1] Batch [312]\tSpeed: 64842.75 samples/sec\tRMSE=3.695260\n",
      "INFO:root:Epoch[1] Batch [624]\tSpeed: 66291.53 samples/sec\tRMSE=3.687865\n",
      "INFO:root:Epoch[1] Batch [936]\tSpeed: 62540.87 samples/sec\tRMSE=3.656047\n",
      "INFO:root:Epoch[1] Batch [1248]\tSpeed: 67802.81 samples/sec\tRMSE=3.422298\n",
      "INFO:root:Epoch[1] Train-RMSE=3.140008\n",
      "INFO:root:Epoch[1] Time cost=1.233\n",
      "INFO:root:Epoch[1] Validation-RMSE=3.310249\n",
      "INFO:root:Epoch[2] Batch [312]\tSpeed: 64824.02 samples/sec\tRMSE=2.766653\n",
      "INFO:root:Epoch[2] Batch [624]\tSpeed: 64551.38 samples/sec\tRMSE=2.157919\n",
      "INFO:root:Epoch[2] Batch [936]\tSpeed: 63794.90 samples/sec\tRMSE=1.791587\n",
      "INFO:root:Epoch[2] Batch [1248]\tSpeed: 67786.24 samples/sec\tRMSE=1.556971\n",
      "INFO:root:Epoch[2] Train-RMSE=1.483133\n",
      "INFO:root:Epoch[2] Time cost=1.235\n",
      "INFO:root:Epoch[2] Validation-RMSE=1.589926\n",
      "INFO:root:Epoch[3] Batch [312]\tSpeed: 64572.03 samples/sec\tRMSE=1.388250\n",
      "INFO:root:Epoch[3] Batch [624]\tSpeed: 67723.48 samples/sec\tRMSE=1.295879\n",
      "INFO:root:Epoch[3] Batch [936]\tSpeed: 61854.14 samples/sec\tRMSE=1.233358\n",
      "INFO:root:Epoch[3] Batch [1248]\tSpeed: 62523.69 samples/sec\tRMSE=1.189739\n",
      "INFO:root:Epoch[3] Train-RMSE=1.164325\n",
      "INFO:root:Epoch[3] Time cost=1.256\n",
      "INFO:root:Epoch[3] Validation-RMSE=1.252920\n",
      "INFO:root:Epoch[4] Batch [312]\tSpeed: 58991.07 samples/sec\tRMSE=1.133154\n",
      "INFO:root:Epoch[4] Batch [624]\tSpeed: 58834.63 samples/sec\tRMSE=1.108365\n",
      "INFO:root:Epoch[4] Batch [936]\tSpeed: 51809.62 samples/sec\tRMSE=1.092431\n",
      "INFO:root:Epoch[4] Batch [1248]\tSpeed: 68050.28 samples/sec\tRMSE=1.077801\n",
      "INFO:root:Epoch[4] Train-RMSE=1.158075\n",
      "INFO:root:Epoch[4] Time cost=1.367\n",
      "INFO:root:Epoch[4] Validation-RMSE=1.136881\n",
      "INFO:root:Epoch[5] Batch [312]\tSpeed: 63840.18 samples/sec\tRMSE=1.045013\n",
      "INFO:root:Epoch[5] Batch [624]\tSpeed: 54559.02 samples/sec\tRMSE=1.041145\n",
      "INFO:root:Epoch[5] Batch [936]\tSpeed: 56593.49 samples/sec\tRMSE=1.044665\n",
      "INFO:root:Epoch[5] Batch [1248]\tSpeed: 48774.98 samples/sec\tRMSE=1.029294\n",
      "INFO:root:Epoch[5] Train-RMSE=1.075923\n",
      "INFO:root:Epoch[5] Time cost=1.450\n",
      "INFO:root:Epoch[5] Validation-RMSE=1.084058\n",
      "INFO:root:Epoch[6] Batch [312]\tSpeed: 57131.73 samples/sec\tRMSE=1.005409\n",
      "INFO:root:Epoch[6] Batch [624]\tSpeed: 52827.36 samples/sec\tRMSE=1.016771\n",
      "INFO:root:Epoch[6] Batch [936]\tSpeed: 62829.42 samples/sec\tRMSE=1.013194\n",
      "INFO:root:Epoch[6] Batch [1248]\tSpeed: 55852.82 samples/sec\tRMSE=1.006081\n",
      "INFO:root:Epoch[6] Train-RMSE=1.063851\n",
      "INFO:root:Epoch[6] Time cost=1.412\n",
      "INFO:root:Epoch[6] Validation-RMSE=1.054935\n",
      "INFO:root:Epoch[7] Batch [312]\tSpeed: 67856.54 samples/sec\tRMSE=0.987511\n",
      "INFO:root:Epoch[7] Batch [624]\tSpeed: 63437.23 samples/sec\tRMSE=0.997116\n",
      "INFO:root:Epoch[7] Batch [936]\tSpeed: 67572.45 samples/sec\tRMSE=0.991542\n",
      "INFO:root:Epoch[7] Batch [1248]\tSpeed: 63492.25 samples/sec\tRMSE=0.995827\n",
      "INFO:root:Epoch[7] Train-RMSE=0.940500\n",
      "INFO:root:Epoch[7] Time cost=1.228\n",
      "INFO:root:Epoch[7] Validation-RMSE=1.038229\n",
      "INFO:root:Epoch[8] Batch [312]\tSpeed: 65570.33 samples/sec\tRMSE=0.982947\n",
      "INFO:root:Epoch[8] Batch [624]\tSpeed: 63110.60 samples/sec\tRMSE=0.979931\n",
      "INFO:root:Epoch[8] Batch [936]\tSpeed: 64586.87 samples/sec\tRMSE=0.982295\n",
      "INFO:root:Epoch[8] Batch [1248]\tSpeed: 53426.81 samples/sec\tRMSE=0.982971\n",
      "INFO:root:Epoch[8] Train-RMSE=0.953352\n",
      "INFO:root:Epoch[8] Time cost=1.313\n",
      "INFO:root:Epoch[8] Validation-RMSE=1.028586\n",
      "INFO:root:Epoch[9] Batch [312]\tSpeed: 59365.85 samples/sec\tRMSE=0.963168\n",
      "INFO:root:Epoch[9] Batch [624]\tSpeed: 66323.45 samples/sec\tRMSE=0.976033\n",
      "INFO:root:Epoch[9] Batch [936]\tSpeed: 69935.30 samples/sec\tRMSE=0.982855\n",
      "INFO:root:Epoch[9] Batch [1248]\tSpeed: 66512.54 samples/sec\tRMSE=0.974915\n",
      "INFO:root:Epoch[9] Train-RMSE=0.923108\n",
      "INFO:root:Epoch[9] Time cost=1.233\n",
      "INFO:root:Epoch[9] Validation-RMSE=1.019939\n"
     ]
    }
   ],
   "source": [
    "# @@@ AUTOTEST_OUTPUT_IGNORED_CELL\n",
    "def plain_net(k):\n",
    "    # input\n",
    "    user = mx.symbol.Variable('user')\n",
    "    item = mx.symbol.Variable('item')\n",
    "    score = mx.symbol.Variable('score')\n",
    "    # user feature lookup\n",
    "    user = mx.symbol.Embedding(data = user, input_dim = max_user, output_dim = k) \n",
    "    # item feature lookup\n",
    "    item = mx.symbol.Embedding(data = item, input_dim = max_item, output_dim = k)\n",
    "    # predict by the inner product, which is elementwise product and then sum\n",
    "    pred = user * item\n",
    "    pred = mx.symbol.sum_axis(data = pred, axis = 1)\n",
    "    pred = mx.symbol.Flatten(data = pred)\n",
    "    # loss layer\n",
    "    pred = mx.symbol.LinearRegressionOutput(data = pred, label = score)\n",
    "    return pred\n",
    "\n",
    "model = train(plain_net(64), batch_size=64, num_epoch=10, learning_rate=.05)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/base_module.py:65: UserWarning: Data provided by label_shapes don't match names specified by label_names ([] vs. ['score'])\n",
      "  warnings.warn(msg)\n"
     ]
    },
    {
     "ename": "AssertionError",
     "evalue": "Shape of unspecified array arg:score changed. This can cause the new executor to not share parameters with the old one. Please check for error in network.If this is intended, set partial_shaping=True to suppress this warning.",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAssertionError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-15-c6918a3c3dca>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[0musers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandom\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmax_user\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m16\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m     \u001b[0mitems\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandom\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmax_item\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0msize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m16\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m     batch_size=4))\n\u001b[0m",
      "\u001b[0;32m<ipython-input-13-3d209d85c5ce>\u001b[0m in \u001b[0;36mpredict\u001b[0;34m(net, users, items, batch_size)\u001b[0m\n\u001b[1;32m      3\u001b[0m     \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m{\u001b[0m\u001b[0;34m'user'\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0musers\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'item'\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mitems\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m     \u001b[0mdata_iter\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mio\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mNDArrayIter\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mbatch_size\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mnet\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_iter\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/base_module.pyc\u001b[0m in \u001b[0;36mpredict\u001b[0;34m(self, eval_data, num_batch, merge_batches, reset, always_output_list)\u001b[0m\n\u001b[1;32m    350\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mnum_batch\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mNone\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0mnbatch\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mnum_batch\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    351\u001b[0m                 \u001b[0;32mbreak\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 352\u001b[0;31m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0meval_batch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mis_train\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    353\u001b[0m             \u001b[0mpad\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0meval_batch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpad\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    354\u001b[0m             \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0mpad\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcopy\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mout\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_outputs\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/module.pyc\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, data_batch, is_train)\u001b[0m\n\u001b[1;32m    605\u001b[0m                 \u001b[0mnew_lshape\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    606\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 607\u001b[0;31m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnew_dshape\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mnew_lshape\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    608\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    609\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_exec_group\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_batch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mis_train\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/module.pyc\u001b[0m in \u001b[0;36mreshape\u001b[0;34m(self, data_shapes, label_shapes)\u001b[0m\n\u001b[1;32m    467\u001b[0m             self.data_names, self.label_names, data_shapes, label_shapes)\n\u001b[1;32m    468\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 469\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_exec_group\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_data_shapes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_label_shapes\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    470\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    471\u001b[0m     def init_optimizer(self, kvstore='local', optimizer='sgd',\n",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/executor_group.pyc\u001b[0m in \u001b[0;36mreshape\u001b[0;34m(self, data_shapes, label_shapes)\u001b[0m\n\u001b[1;32m    352\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_default_execs\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    353\u001b[0m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_default_execs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexecs\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 354\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbind_exec\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata_shapes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabel_shapes\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mreshape\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    355\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    356\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mset_params\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0marg_params\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maux_params\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_extra\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/module/executor_group.pyc\u001b[0m in \u001b[0;36mbind_exec\u001b[0;34m(self, data_shapes, label_shapes, shared_group, reshape)\u001b[0m\n\u001b[1;32m    328\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mreshape\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    329\u001b[0m                 self.execs[i] = self._default_execs[i].reshape(\n\u001b[0;32m--> 330\u001b[0;31m                     allow_up_sizing=True, **dict(data_shapes_i + label_shapes_i))\n\u001b[0m\u001b[1;32m    331\u001b[0m             \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    332\u001b[0m                 self.execs.append(self._bind_ith_exec(i, data_shapes_i, label_shapes_i,\n",
      "\u001b[0;32m/path/to/.local/share/virtualenvs/mxnet/lib/python2.7/site-packages/mxnet/executor.pyc\u001b[0m in \u001b[0;36mreshape\u001b[0;34m(self, partial_shaping, allow_up_sizing, **kwargs)\u001b[0m\n\u001b[1;32m    428\u001b[0m                     \u001b[0;34m\"This can cause the new executor to not share parameters \"\u001b[0m \u001b[0;34m+\u001b[0m\u001b[0;31m \u001b[0m\u001b[0;31m\\\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    429\u001b[0m                     \u001b[0;34m\"with the old one. Please check for error in network.\"\u001b[0m \u001b[0;34m+\u001b[0m\u001b[0;31m\\\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 430\u001b[0;31m                     \"If this is intended, set partial_shaping=True to suppress this warning.\")\n\u001b[0m\u001b[1;32m    431\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    432\u001b[0m         \u001b[0mnew_aux_dict\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m{\u001b[0m\u001b[0;34m}\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mAssertionError\u001b[0m: Shape of unspecified array arg:score changed. This can cause the new executor to not share parameters with the old one. Please check for error in network.If this is intended, set partial_shaping=True to suppress this warning."
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "print(predict(\n",
    "    model,\n",
    "    users=np.random.randint(0, max_user, size=(16,)),\n",
    "    items=np.random.randint(0, max_item, size=(16,)),\n",
    "    batch_size=4))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}

Thanks

8

I didn’t realize you switched to using the updated tutorial with Module interface. The code that I provided works fine with the deprecated FeedForward. The Module interface performs a few more sanity checks and that’s why your call fails.

What does it mean that Shape of unspecified array changed?
What is happening here is that the underlying model has initialized item, user, and score variables based on the data and label shape used in train() and that shape has a batch_size of 64. However in predict example I provided, I use batch_size of 16. The model does notice that you’re explicitly changing the shape of user and item variables, but you are not explicit about changing shape of score variable, so it complains that you must be explicit. A few ways to fix this:

  1. Call reshape() with new batch_size parameter (e.g. 16) before predict() :

    model.reshape([('user', (16,)), ('item', (16,))], [('score', (16,))])

  2. Call predict() with the same batch_size as training:

    print(predict(
    model,
    users=np.random.randint(0, max_user, size=(64,)),
    items=np.random.randint(0, max_item, size=(64,)),
    batch_size=64))

  3. If you pay attention, you’ll notice that providing a size for score is redundant for prediction. However because your model was created to include a loss, you always have to specify a size for that variable. Alternatively, you can create a new model for predict that doesn’t include the loss. The following changes can be made to plain_net() to create a model for training or prediction (notice that I’m also passing a name to mx.symbol.Embedding to keep the arguments names consistent between the training and prediction models:

def plain_net(k, max_user, max_item, is_training = True):
    # input
    user = mx.symbol.Variable('user')
    item = mx.symbol.Variable('item')
    score = mx.symbol.Variable('score')
    # user feature lookup
    user = mx.symbol.Embedding(data = user, input_dim = max_user, output_dim = k, name='user_embed')
    # item feature lookup
    item = mx.symbol.Embedding(data = item, input_dim = max_item, output_dim = k, name='item_embed')
    # predict by the inner product, which is elementwise product and then sum
    pred = user * item
    pred = mx.symbol.sum_axis(data = pred, axis = 1)
    pred = mx.symbol.Flatten(data = pred)
    if is_training:
        # loss layer
        pred = mx.symbol.LinearRegressionOutput(data=pred, label=score)
    return pred

Then you can predict as follows:

    users = np.random.randint(0, num_users, size=(16,))
    items = np.random.randint(0, num_items, size=(16,))

    data = {'user': users, 'item': items}
    data_iter = mx.io.NDArrayIter(data, batch_size=4)

    model_pred = mx.mod.Module(
        symbol=plain_net(64, num_users, num_items, is_training=False),
        data_names=[x[0] for x in data_iter.provide_data],
        context=[mx.cpu()])
    model_pred.bind(data_shapes=data_iter.provide_data)

    #Read parameter from trained model and load them into the prediction model
    arg_params, aux_params = model.get_params()
    model_pred.set_params(arg_params, aux_params)
    # Preform prediction
    print(model_pred.predict(data_iter))
9

Happy New Year! Sorry for the late reply. I was off the grid for a few days.

I’m able to get results with option 3, thanks! One last question, when I run the prediction for 2 users and 2 items like this:

users = np.random.randint(0, max_user, size=(2,))
items = np.random.randint(0, max_item, size=(2,))

Then I get back an NDArray 2x1, for example [[ 2.40511107] [ 3.21094108]]. I actually would have expected to get 4 values back, like:

  • user 1 likely to rate movie 1 this much
  • user 1 likely to rate movie 2 this much
  • user 2 likely to rate movie 1 this much
  • user 2 likely to rate movie 2 this much

Likewise when I predict the rating for 1 item for 10 users, I would expect to get 10 different results. But I only get one value back…

10

For each prediction, the network needs a value for each variable, i.e. user and item variables. Think of it as a pipeline of (user, item) pairs of input data. The output[0] prediction is for (users[0], items[0]) pair and the output[1] prediction is for (users[1], items[1]) pair.

11

Got it, that makes total sense. Thanks again for your help!

12

Hi,
I’m at the beginning with mxnet, I’m working on this same code (with modules), and I did a quick manual test to verify the error in predictions. But I found something strange.

I checked on the dataset u1.base that the user 1 scored with “5”, the movies: 1,13,15 and with “1” the movies: 8,21,29.

users=np.full((1,64),1)[0]
items=np.array([1,13,15,8,21,29,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
           1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1])

pred=predict(
   model,
   users=users,
   items=items,
   batch_size=64)

the results is:

[[ 3.96710515]
 [ 3.61739969]
 [ 3.60116005]
 [ 3.96544552]
 [ 3.50047493]
 [ 4.00641775]

…and it is quite far from the original score [5,5,5,1,1,1]. But the problem is also that if I test the prediction with a different user_id, I will receive almost the same answer.
Where is the error in the procedure?

13

There are three example networks, plain_net, one_layer_mlp, and one_layer_dropout_mlp. Which one did you use in your prediction?