make lesson2 runnable

.travis.yml

@@ -3,6 +3,10 @@ python: 2.7
 dist: trusty
 sudo: false
 
+env:
+  global:
+    - NBCONVERT_ARGS="--to html --execute --ExecutePreprocessor.kernel_name=python --ExecutePreprocessor.timeout=-1"
+
 install:
   # xvfb
   # https://docs.travis-ci.com/user/gui-and-headless-browsers/
@@ -32,7 +36,8 @@ script:
     curl -sSLO "http://files.fast.ai/data/dogscats.zip" &&
     unzip -qq -d deeplearning1/nbs/data dogscats.zip;
     fi
-  - jupyter nbconvert --to html --execute deeplearning1/nbs/lesson1.ipynb --ExecutePreprocessor.kernel_name=python --ExecutePreprocessor.timeout=-1
+  - travis_wait jupyter nbconvert deeplearning1/nbs/lesson1.ipynb $NBCONVERT_ARGS
+  - travis_wait jupyter nbconvert deeplearning1/nbs/lesson2.ipynb $NBCONVERT_ARGS
 
 before_cache:
   # We only cache the archives but not the extracted packages,

deeplearning1/nbs/lesson2.ipynb

@@ -374,8 +374,8 @@
    },
    "outputs": [],
    "source": [
-    "#path = \"data/dogscats/sample/\"\n",
-    "path = \"data/dogscats/\"\n",
+    "path = \"data/dogscats/sample/\"\n",
+    "# path = \"data/dogscats/\"\n",
     "model_path = path + 'models/'\n",
     "if not os.path.exists(model_path): os.mkdir(model_path)"
    ]
@@ -395,8 +395,8 @@
    },
    "outputs": [],
    "source": [
-    "batch_size=100\n",
-    "#batch_size=4"
+    "# batch_size=100\n",
+    "batch_size=4"
    ]
   },
   {
@@ -1160,7 +1160,10 @@
     "#3. The images we were most confident were cats, but are actually dogs\n",
     "incorrect_cats = np.where((preds==0) & (preds!=val_labels[:,1]))[0]\n",
     "most_incorrect_cats = np.argsort(probs[incorrect_cats])[::-1][:n_view]\n",
-    "plots_idx(incorrect_cats[most_incorrect_cats], probs[incorrect_cats][most_incorrect_cats])"
+    "if len(most_incorrect_cats):\n",
+    "    plots_idx(incorrect_cats[most_incorrect_cats], probs[incorrect_cats][most_incorrect_cats])\n",
+    "else:\n",
+    "    print('No incorrect cats!')"
    ]
   },
   {
@@ -1187,7 +1190,10 @@
     "#3. The images we were most confident were dogs, but are actually cats\n",
     "incorrect_dogs = np.where((preds==1) & (preds!=val_labels[:,1]))[0]\n",
     "most_incorrect_dogs = np.argsort(probs[incorrect_dogs])[:n_view]\n",
-    "plots_idx(incorrect_dogs[most_incorrect_dogs], 1-probs[incorrect_dogs][most_incorrect_dogs])"
+    "if len(most_incorrect_dogs):\n",
+    "    plots_idx(incorrect_dogs[most_incorrect_dogs], 1-probs[incorrect_dogs][most_incorrect_dogs])\n",
+    "else:\n",
+    "    print('No incorrect dogs!')"
    ]
   },
   {
@@ -1519,8 +1525,8 @@
    "outputs": [],
    "source": [
     "def fit_model(model, batches, val_batches, nb_epoch=1):\n",
-    "    model.fit_generator(batches, samples_per_epoch=batches.N, nb_epoch=nb_epoch, \n",
-    "                        validation_data=val_batches, nb_val_samples=val_batches.N)"
+    "    model.fit_generator(batches, samples_per_epoch=batches.n, nb_epoch=nb_epoch, \n",
+    "                        validation_data=val_batches, nb_val_samples=val_batches.n)"
    ]
   },
   {
@@ -1958,7 +1964,7 @@
    "outputs": [],
    "source": [
     "model.load_weights(model_path+'finetune2.h5')\n",
-    "model.evaluate_generator(get_batches(path+'valid', gen, False, batch_size*2), val_batches.N)"
+    "model.evaluate_generator(get_batches(path+'valid', gen, False, batch_size*2), val_batches.n)"
    ]
   },
   {

requirements.txt

@@ -6,4 +6,5 @@ theano
 tensorflow
 pandas
 scikit-learn
-bcolz
+bcolz
+sympy