Feature/torch 0.3 (#127)

* Better error message when data conversion fails. Specifically, when conversion to torch tensor or to numpy array fails, users would often only get "AttributeError: bla has no attribute .cuda". Now a more specific TypeError is shown. * Update install instructions for pytorch 0.3. * Softmax now requires a dimension parameter. * Fix an sklearn deprecation warning. * Install pytorch from new pytorch forge channel. * Update install instructions for pytorch 0.3. * Softmax now requires a dimension parameter. * Fix an sklearn deprecation warning. * Install pytorch from new pytorch forge channel. * Add missing parameter to a Scoring unit test. * __repr__ changed in torch 0.3, adjust test to changes. * Remove temporary workaround to pytorch indexing idiosyncrasy. * Fix some pylint complaints.
skorch-dev · Dec 8, 2017 · d2fcce0 · d2fcce0
1 parent 4bf56dd
commit d2fcce0
Show file tree

Hide file tree

Showing 11 changed files with 237 additions and 228 deletions.
diff --git a/.travis.yml b/.travis.yml
@@ -21,6 +21,6 @@ install:
   - source activate skorch-env
   - conda install --file=requirements-dev.txt
   - python setup.py install
-  - conda install -c soumith pytorch
+  - conda install -c pytorch pytorch
 script:
   - pytest
diff --git a/README.rst b/README.rst
@@ -195,11 +195,11 @@ version you need is dependent on your system. For installation
 instructions for ``pytorch``, visit the `pytorch website
 <http://pytorch.org/>`__.
 
-In general, this should work:
+In general, this should work (assuming CUDA 9):
 
 .. code:: bash
 
     # using conda:
-    conda install pytorch cuda80 -c soumith
+    conda install pytorch cuda90 -c pytorch
     # using pip
-    pip install http://download.pytorch.org/whl/cu80/torch-0.2.0.post3-cp36-cp36m-manylinux1_x86_64.whl
+    pip install http://download.pytorch.org/whl/cu90/torch-0.3.0.post4-cp36-cp36m-linux_x86_64.whl
diff --git a/notebooks/Advanced_Usage.ipynb b/notebooks/Advanced_Usage.ipynb
@@ -177,7 +177,7 @@
     "        X = self.nonlin(self.dense0(X))\n",
     "        X = self.dropout(X)\n",
     "        X = F.relu(self.dense1(X))\n",
-    "        X = F.softmax(self.output(X))\n",
+    "        X = F.softmax(self.output(X), dim=-1)\n",
     "        return X"
    ]
   },
@@ -283,7 +283,7 @@
     "net = NeuralNetClassifier(\n",
     "    ClassifierModule,\n",
     "    max_epochs=10,\n",
-    "    lr=0.1,\n",
+    "    lr=0.02,\n",
     "    warm_start=True,\n",
     "    callbacks=[AccuracyTweet(min_accuracy=0.7)],\n",
     ")"
@@ -300,16 +300,16 @@
      "text": [
       "  epoch    train_loss    valid_acc    valid_loss     dur\n",
       "-------  ------------  -----------  ------------  ------\n",
-      "      1        \u001b[36m0.7111\u001b[0m       \u001b[32m0.5100\u001b[0m        \u001b[35m0.6894\u001b[0m  0.0994\n",
-      "      2        \u001b[36m0.6928\u001b[0m       \u001b[32m0.5500\u001b[0m        \u001b[35m0.6803\u001b[0m  0.0676\n",
-      "      3        \u001b[36m0.6833\u001b[0m       \u001b[32m0.5650\u001b[0m        \u001b[35m0.6741\u001b[0m  0.0396\n",
-      "      4        \u001b[36m0.6763\u001b[0m       \u001b[32m0.5850\u001b[0m        \u001b[35m0.6674\u001b[0m  0.0661\n",
-      "      5        \u001b[36m0.6727\u001b[0m       \u001b[32m0.6450\u001b[0m        \u001b[35m0.6616\u001b[0m  0.0557\n",
-      "      6        \u001b[36m0.6606\u001b[0m       \u001b[32m0.6600\u001b[0m        \u001b[35m0.6536\u001b[0m  0.0645\n",
-      "      7        \u001b[36m0.6560\u001b[0m       0.6600        \u001b[35m0.6443\u001b[0m  0.0559\n",
-      "      8        \u001b[36m0.6427\u001b[0m       \u001b[32m0.6650\u001b[0m        \u001b[35m0.6354\u001b[0m  0.0569\n",
-      "      9        \u001b[36m0.6300\u001b[0m       \u001b[32m0.6800\u001b[0m        \u001b[35m0.6264\u001b[0m  0.0606\n",
-      "     10        \u001b[36m0.6289\u001b[0m       0.6800        \u001b[35m0.6189\u001b[0m  0.0594\n",
+      "      1        \u001b[36m0.6908\u001b[0m       \u001b[32m0.5950\u001b[0m        \u001b[35m0.6842\u001b[0m  0.1078\n",
+      "      2        \u001b[36m0.6876\u001b[0m       0.5950        \u001b[35m0.6815\u001b[0m  0.0561\n",
+      "      3        \u001b[36m0.6853\u001b[0m       \u001b[32m0.6100\u001b[0m        \u001b[35m0.6789\u001b[0m  0.0580\n",
+      "      4        0.6882       0.5950        \u001b[35m0.6769\u001b[0m  0.0551\n",
+      "      5        \u001b[36m0.6780\u001b[0m       0.6000        \u001b[35m0.6743\u001b[0m  0.0570\n",
+      "      6        \u001b[36m0.6730\u001b[0m       0.6100        \u001b[35m0.6717\u001b[0m  0.0538\n",
+      "      7        \u001b[36m0.6664\u001b[0m       \u001b[32m0.6150\u001b[0m        \u001b[35m0.6698\u001b[0m  0.0496\n",
+      "      8        0.6670       0.6100        \u001b[35m0.6670\u001b[0m  0.0522\n",
+      "      9        0.6667       \u001b[32m0.6300\u001b[0m        \u001b[35m0.6646\u001b[0m  0.0560\n",
+      "     10        \u001b[36m0.6624\u001b[0m       \u001b[32m0.6350\u001b[0m        \u001b[35m0.6624\u001b[0m  0.0575\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n",
       "*tweet* Accuracy never reached 0.7 :( #skorch #pytorch\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n"
@@ -318,7 +318,7 @@
     {
      "data": {
       "text/plain": [
-       "<skorch.net.NeuralNetClassifier at 0x7f7aeeae95f8>"
+       "<skorch.net.NeuralNetClassifier at 0x7faf55866048>"
       ]
      },
      "execution_count": 10,
@@ -346,25 +346,25 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "     11        \u001b[36m0.6241\u001b[0m       \u001b[32m0.7150\u001b[0m        \u001b[35m0.6114\u001b[0m  0.0475\n",
-      "     12        \u001b[36m0.6132\u001b[0m       0.7150        \u001b[35m0.6017\u001b[0m  0.0443\n",
-      "     13        \u001b[36m0.5950\u001b[0m       \u001b[32m0.7350\u001b[0m        \u001b[35m0.5902\u001b[0m  0.0523\n",
-      "     14        \u001b[36m0.5914\u001b[0m       0.7200        \u001b[35m0.5831\u001b[0m  0.0551\n",
-      "     15        \u001b[36m0.5784\u001b[0m       0.7300        \u001b[35m0.5733\u001b[0m  0.0598\n",
-      "     16        0.5816       \u001b[32m0.7400\u001b[0m        \u001b[35m0.5665\u001b[0m  0.0524\n",
-      "     17        \u001b[36m0.5766\u001b[0m       \u001b[32m0.7450\u001b[0m        \u001b[35m0.5616\u001b[0m  0.0539\n",
-      "     18        \u001b[36m0.5636\u001b[0m       0.7450        \u001b[35m0.5559\u001b[0m  0.0548\n",
-      "     19        \u001b[36m0.5517\u001b[0m       0.7350        \u001b[35m0.5527\u001b[0m  0.0571\n",
-      "     20        0.5570       0.7350        \u001b[35m0.5492\u001b[0m  0.0544\n",
+      "     11        0.6647       \u001b[32m0.6500\u001b[0m        \u001b[35m0.6598\u001b[0m  0.0816\n",
+      "     12        \u001b[36m0.6573\u001b[0m       \u001b[32m0.6650\u001b[0m        \u001b[35m0.6575\u001b[0m  0.0675\n",
+      "     13        \u001b[36m0.6458\u001b[0m       \u001b[32m0.6700\u001b[0m        \u001b[35m0.6549\u001b[0m  0.0454\n",
+      "     14        0.6528       \u001b[32m0.6750\u001b[0m        \u001b[35m0.6525\u001b[0m  0.0905\n",
+      "     15        0.6476       0.6700        \u001b[35m0.6502\u001b[0m  0.0854\n",
+      "     16        0.6483       0.6750        \u001b[35m0.6476\u001b[0m  0.0878\n",
+      "     17        0.6514       \u001b[32m0.6800\u001b[0m        \u001b[35m0.6452\u001b[0m  0.0741\n",
+      "     18        \u001b[36m0.6365\u001b[0m       \u001b[32m0.6850\u001b[0m        \u001b[35m0.6422\u001b[0m  0.0545\n",
+      "     19        \u001b[36m0.6335\u001b[0m       \u001b[32m0.7000\u001b[0m        \u001b[35m0.6390\u001b[0m  0.0531\n",
+      "     20        0.6381       \u001b[32m0.7100\u001b[0m        \u001b[35m0.6363\u001b[0m  0.0477\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n",
-      "*tweet* Accuracy reached 0.7 at epoch 11!!! #skorch #pytorch\n",
+      "*tweet* Accuracy reached 0.7 at epoch 19!!! #skorch #pytorch\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n"
      ]
     },
     {
      "data": {
       "text/plain": [
-       "<skorch.net.NeuralNetClassifier at 0x7f7aeeae95f8>"
+       "<skorch.net.NeuralNetClassifier at 0x7faf55866048>"
       ]
      },
      "execution_count": 11,
@@ -460,25 +460,25 @@
      "text": [
       "  epoch    train_loss    valid_acc    valid_loss     dur\n",
       "-------  ------------  -----------  ------------  ------\n",
-      "      1        \u001b[36m0.7261\u001b[0m       \u001b[32m0.5050\u001b[0m        \u001b[35m0.6986\u001b[0m  0.0527\n",
-      "      2        \u001b[36m0.6977\u001b[0m       \u001b[32m0.5350\u001b[0m        \u001b[35m0.6889\u001b[0m  0.0584\n",
-      "      3        \u001b[36m0.6897\u001b[0m       \u001b[32m0.5550\u001b[0m        \u001b[35m0.6844\u001b[0m  0.0593\n",
-      "      4        \u001b[36m0.6846\u001b[0m       \u001b[32m0.5800\u001b[0m        \u001b[35m0.6813\u001b[0m  0.0579\n",
-      "      5        \u001b[36m0.6788\u001b[0m       0.5800        \u001b[35m0.6768\u001b[0m  0.0483\n",
-      "      6        \u001b[36m0.6725\u001b[0m       0.5800        \u001b[35m0.6731\u001b[0m  0.0479\n",
-      "      7        \u001b[36m0.6711\u001b[0m       \u001b[32m0.5950\u001b[0m        \u001b[35m0.6689\u001b[0m  0.0522\n",
-      "      8        \u001b[36m0.6581\u001b[0m       \u001b[32m0.6150\u001b[0m        \u001b[35m0.6650\u001b[0m  0.0542\n",
-      "      9        0.6648       0.6050        \u001b[35m0.6605\u001b[0m  0.0585\n",
-      "     10        \u001b[36m0.6550\u001b[0m       0.6150        \u001b[35m0.6549\u001b[0m  0.0544\n",
+      "      1        \u001b[36m0.7139\u001b[0m       \u001b[32m0.5500\u001b[0m        \u001b[35m0.6933\u001b[0m  0.1162\n",
+      "      2        \u001b[36m0.6916\u001b[0m       \u001b[32m0.5950\u001b[0m        \u001b[35m0.6873\u001b[0m  0.0457\n",
+      "      3        \u001b[36m0.6829\u001b[0m       0.5800        \u001b[35m0.6814\u001b[0m  0.0516\n",
+      "      4        \u001b[36m0.6671\u001b[0m       \u001b[32m0.6050\u001b[0m        \u001b[35m0.6718\u001b[0m  0.0440\n",
+      "      5        \u001b[36m0.6670\u001b[0m       \u001b[32m0.6200\u001b[0m        \u001b[35m0.6639\u001b[0m  0.1035\n",
+      "      6        \u001b[36m0.6622\u001b[0m       \u001b[32m0.6350\u001b[0m        \u001b[35m0.6546\u001b[0m  0.0561\n",
+      "      7        \u001b[36m0.6371\u001b[0m       \u001b[32m0.6550\u001b[0m        \u001b[35m0.6429\u001b[0m  0.0387\n",
+      "      8        \u001b[36m0.6293\u001b[0m       \u001b[32m0.6700\u001b[0m        \u001b[35m0.6312\u001b[0m  0.0680\n",
+      "      9        \u001b[36m0.6170\u001b[0m       0.6650        \u001b[35m0.6200\u001b[0m  0.0500\n",
+      "     10        0.6204       \u001b[32m0.6750\u001b[0m        \u001b[35m0.6119\u001b[0m  0.0477\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n",
-      "*tweet* Accuracy reached 0.6 at epoch 8!!! #skorch #pytorch\n",
+      "*tweet* Accuracy reached 0.6 at epoch 4!!! #skorch #pytorch\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n"
      ]
     },
     {
      "data": {
       "text/plain": [
-       "<skorch.net.NeuralNetClassifier at 0x7f7aeeae94e0>"
+       "<skorch.net.NeuralNetClassifier at 0x7faf5585b9b0>"
       ]
      },
      "execution_count": 13,
@@ -505,7 +505,7 @@
     {
      "data": {
       "text/plain": [
-       "<skorch.net.NeuralNetClassifier at 0x7f7aeeae94e0>"
+       "<skorch.net.NeuralNetClassifier at 0x7faf5585b9b0>"
       ]
      },
      "execution_count": 14,
@@ -528,16 +528,16 @@
      "text": [
       "  epoch    train_loss    valid_acc    valid_loss     dur\n",
       "-------  ------------  -----------  ------------  ------\n",
-      "     11        \u001b[36m0.6435\u001b[0m       \u001b[32m0.6250\u001b[0m        \u001b[35m0.6492\u001b[0m  0.0568\n",
-      "     12        0.6435       \u001b[32m0.6350\u001b[0m        \u001b[35m0.6437\u001b[0m  0.0546\n",
-      "     13        \u001b[36m0.6267\u001b[0m       \u001b[32m0.6450\u001b[0m        \u001b[35m0.6375\u001b[0m  0.0555\n",
-      "     14        \u001b[36m0.6214\u001b[0m       \u001b[32m0.6800\u001b[0m        \u001b[35m0.6306\u001b[0m  0.0390\n",
-      "     15        \u001b[36m0.6185\u001b[0m       0.6750        \u001b[35m0.6239\u001b[0m  0.0440\n",
-      "     16        \u001b[36m0.6060\u001b[0m       0.6750        \u001b[35m0.6154\u001b[0m  0.0454\n",
-      "     17        \u001b[36m0.5964\u001b[0m       \u001b[32m0.6850\u001b[0m        \u001b[35m0.6061\u001b[0m  0.0503\n",
-      "     18        \u001b[36m0.5868\u001b[0m       \u001b[32m0.7000\u001b[0m        \u001b[35m0.5964\u001b[0m  0.0579\n",
-      "     19        \u001b[36m0.5693\u001b[0m       \u001b[32m0.7150\u001b[0m        \u001b[35m0.5859\u001b[0m  0.0556\n",
-      "     20        \u001b[36m0.5689\u001b[0m       \u001b[32m0.7200\u001b[0m        \u001b[35m0.5793\u001b[0m  0.0558\n",
+      "     11        \u001b[36m0.5845\u001b[0m       \u001b[32m0.7000\u001b[0m        \u001b[35m0.6016\u001b[0m  0.0998\n",
+      "     12        \u001b[36m0.5831\u001b[0m       \u001b[32m0.7050\u001b[0m        \u001b[35m0.5915\u001b[0m  0.0762\n",
+      "     13        0.5854       \u001b[32m0.7200\u001b[0m        \u001b[35m0.5788\u001b[0m  0.0597\n",
+      "     14        \u001b[36m0.5582\u001b[0m       0.7150        \u001b[35m0.5729\u001b[0m  0.0520\n",
+      "     15        0.5601       0.7150        \u001b[35m0.5692\u001b[0m  0.0486\n",
+      "     16        \u001b[36m0.5468\u001b[0m       \u001b[32m0.7250\u001b[0m        \u001b[35m0.5662\u001b[0m  0.0479\n",
+      "     17        \u001b[36m0.5333\u001b[0m       \u001b[32m0.7300\u001b[0m        \u001b[35m0.5583\u001b[0m  0.0576\n",
+      "     18        0.5592       0.7200        \u001b[35m0.5555\u001b[0m  0.0493\n",
+      "     19        \u001b[36m0.5295\u001b[0m       0.7300        \u001b[35m0.5488\u001b[0m  0.0495\n",
+      "     20        \u001b[36m0.5232\u001b[0m       0.7300        \u001b[35m0.5428\u001b[0m  0.0532\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n",
       "*tweet* Accuracy never reached 0.75 :( #skorch #pytorch\n",
       "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n"
@@ -546,7 +546,7 @@
     {
      "data": {
       "text/plain": [
-       "<skorch.net.NeuralNetClassifier at 0x7f7aeeae94e0>"
+       "<skorch.net.NeuralNetClassifier at 0x7faf5585b9b0>"
       ]
      },
      "execution_count": 15,