Remove fastai tests using 1.x api

Author: rosbo

Dockerfile

@@ -432,6 +432,7 @@ RUN pip install bcolz && \
     # Don't install dependencies for fastai because it requires pytorch<0.4.
     # which downgrades pytorch. fastai does work with pytorch 0.4.
     pip install fastai==0.7.0 --no-deps && \
+    pip install torchtext && \
     # clean up pip cache
     rm -rf /root/.cache/pip/* && \
     cd && rm -rf /usr/local/src/*

tests/test_fastai.py

@@ -4,35 +4,40 @@
 import pandas as pd
 import torch
 
-from fastai.tabular import tabular_data_from_df, get_tabular_learner
 from fastai.core import partition
-from fastai.torch_core import tensor
+from fastai.layer_optimizer import LayerOptimizer
 
 class TestFastAI(unittest.TestCase):
     def test_partition(self):
         result = partition([1,2,3,4,5], 2)
 
         self.assertEqual(3, len(result))
 
-    def test_has_version(self):
-        self.assertGreater(len(fastai.__version__), 1)
-    
-    # based on https://github.com/fastai/fastai/blob/master/tests/test_torch_core.py#L17
-    def test_torch_tensor(self):
-        a = tensor([1, 2, 3])
-        b = torch.tensor([1, 2, 3])
+    # based on https://github.com/fastai/fastai/blob/0.7.0/tests/test_layer_optimizer.py
+    def test_layer_optimizer(self):
+        lo = LayerOptimizer(FakeOpt, fastai_params_('A', 'B', 'C'), 1e-2, 1e-4)
+        fast_check_optimizer_(lo.opt, [(nm, 1e-2, 1e-4) for nm in 'ABC'])
 
-        self.assertTrue(torch.all(a == b))
 
-    def test_tabular(self):
-        df = pd.read_csv("/input/tests/data/train.csv")
+class Par(object):
+    def __init__(self, x, grad=True):
+        self.x = x
+        self.requires_grad = grad
+    def parameters(self): return [self]
 
-        train_df, valid_df = df[:-5].copy(),df[-5:].copy()
-        dep_var = "label"
-        cont_names = []
-        for i in range(784):
-            cont_names.append("pixel" + str(i))
 
-        data = tabular_data_from_df("", train_df, valid_df, dep_var, cont_names=cont_names, cat_names=[])
-        learn = get_tabular_learner(data, layers=[200, 100])
-        learn.fit(epochs=1)
+class FakeOpt(object):
+    def __init__(self, params): self.param_groups = params
+
+
+def fastai_params_(*names): return [Par(nm) for nm in names]
+
+def fast_check_optimizer_(opt, expected):
+    actual = opt.param_groups
+    assert len(actual) == len(expected)
+    for (a, e) in zip(actual, expected): fastai_check_param_(a, *e)
+
+def fastai_check_param_(par, nm, lr, wd):
+    assert par['params'][0].x == nm
+    assert par['lr'] == lr
+    assert par['weight_decay'] == wd