fastai v1

dev_nb/009_rossmann.ipynb

@@ -16,6 +16,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "#export\n",
     "from nb_008 import *"
    ]
   },
@@ -498,6 +499,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "#export\n",
     "ListSizes = Collection[Tuple[int,int]]\n",
     "OptRange = Optional[Tuple[float,float]]\n",
     "\n",

dev_nb/Untitled.ipynb

@@ -0,0 +1,38 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from nb_009 import *"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "ItemsList"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

dev_nb/nb_006c.py → dev_nb/nb_001a.py

@@ -2,7 +2,4 @@
         #################################################
         ### THIS FILE WAS AUTOGENERATED! DO NOT EDIT! ###
         #################################################
-        # file to edit: dev_nb/006c_camvid.ipynb
-
-from nb_006a import *
-from concurrent.futures import ProcessPoolExecutor
+        # file to edit: dev_nb/001a_nn_basics.ipynb

dev_nb/nb_006a.py

@@ -7,12 +7,16 @@
 from nb_006 import *
 import gc
 
-def in_channels(m):
+Sizes = List[List[int]]
+
+def in_channels(m:Model) -> List[int]:
+    "Returns the shape of the first weight layer"
     for l in flatten_model(m):
         if hasattr(l, 'weight'): return l.weight.shape[1]
     raise Exception('No weight layer')
 
-def model_sizes(m, size=(256,256), full=True):
+def model_sizes(m:Model, size:tuple=(256,256), full:bool=True) -> Tuple[Sizes,Tensor,Hooks]:
+    "Passes a dummy input through the model to get the various sizes"
     hooks = hook_outputs(m)
     ch_in = in_channels(m)
     x = torch.zeros(1,ch_in,*size)
@@ -21,7 +25,8 @@ def model_sizes(m, size=(256,256), full=True):
     if not full: hooks.remove()
     return res,x,hooks if full else res
 
-def get_sfs_idxs(sizes, last=True):
+def get_sfs_idxs(sizes:Sizes, last:bool=True) -> List[int]:
+    "Get the indexes of the layers where the size of the activation changes"
     if last:
         feature_szs = [size[-1] for size in sizes]
         sfs_idxs = list(np.where(np.array(feature_szs[:-1]) != np.array(feature_szs[1:]))[0])
@@ -30,7 +35,8 @@ def get_sfs_idxs(sizes, last=True):
     return sfs_idxs
 
 class UnetBlock(nn.Module):
-    def __init__(self, up_in_c, x_in_c, hook):
+    "An basic unet block"
+    def __init__(self, up_in_c:int, x_in_c:int, hook:Hook):
         super().__init__()
         self.hook = hook
         ni = up_in_c
@@ -41,15 +47,16 @@ def __init__(self, up_in_c, x_in_c, hook):
         self.conv2 = conv2d(ni, ni)
         self.bn = nn.BatchNorm2d(ni)
 
-    def forward(self, up_in):
+    def forward(self, up_in:Tensor) -> Tensor:
         up_out = self.upconv(up_in)
         cat_x = torch.cat([up_out, self.hook.stored], dim=1)
         x = F.relu(self.conv1(cat_x))
         x = F.relu(self.conv2(x))
         return self.bn(x)
 
 class DynamicUnet(nn.Sequential):
-    def __init__(self, encoder, n_classes, last=True):
+    "Unet created from a given architecture"
+    def __init__(self, encoder:Model, n_classes:int, last:bool=True):
         imsize = (256,256)
         sfs_szs,x,self.sfs = model_sizes(encoder, size=imsize)
         sfs_idxs = reversed(get_sfs_idxs(sfs_szs, last))

dev_nb/nb_007b.py

@@ -76,7 +76,7 @@ def pad_collate(samples:BatchSamples, pad_idx:int=1, pad_first:bool=True) -> Tup
     return res, LongTensor([s[1] for s in samples]).squeeze()
 
 def classifier_data(datasets:Collection[TextDataset], path:PathOrStr, **kwargs) -> DataBunch:
-    "Function that transform the datasets in a DataBunch for classification"
+    "Function that transform the `datasets` in a `DataBunch` for classification"
     bs = kwargs.pop('bs') if 'bs' in kwargs else 64
     pad_idx = kwargs.pop('pad_idx') if 'pad_idx' in kwargs else 1
     train_sampler = SortishSampler(datasets[0].ids, key=lambda x: len(datasets[0].ids[x]), bs=bs//2)
@@ -184,7 +184,7 @@ def load_pretrained(self, wgts_fname:str, itos_fname:str):
     def language_model(cls, data:DataBunch, bptt:int=70, emb_sz:int=400, nh:int=1150, nl:int=3, pad_token:int=1,
                        drop_mult:float=1., tie_weights:bool=True, bias:bool=True, qrnn:bool=False,
                        pretrained_fnames:OptStrTuple=None, **kwargs) -> 'RNNLearner':
-        "Creates a Learner with a language model."
+        "Creates a `Learner` with a language model."
         dps = np.array([0.25, 0.1, 0.2, 0.02, 0.15]) * drop_mult
         vocab_size = len(data.train_ds.vocab.itos)
         model = get_language_model(vocab_size, emb_sz, nh, nl, pad_token, input_p=dps[0], output_p=dps[1],

dev_nb/nb_009.py

@@ -4,6 +4,8 @@
         #################################################
         # file to edit: dev_nb/009_rossmann.ipynb
 
+from nb_008 import *
+
 StrList = Collection[str]
 @dataclass
 class TabularTransform():

fastai/__init__.py

@@ -6,4 +6,7 @@
 from .metrics import *
 from .torch_core import *
 from .train import *
+from .rnn_learner import *
+from .conv_learner import *
+from .colab import *