callbacks/callback_fns tests

tests/test_callback.py

@@ -1,3 +1,83 @@
-import pytest,fastai
+import pytest, fastai
+from fastai.vision import *
+from io import StringIO
+from contextlib import redirect_stdout
+from utils.fakes import fake_data
+from utils.text import apply_print_resets
 
-#I'm all empty, please fill me with tests about the functions in callback.py
+n_in, n_out = 3, 2
+@pytest.fixture(scope="module")
+def data(): return fake_data(n_in=n_in, n_out=n_out)
+
+@pytest.fixture(scope="module")
+def model(): return nn.Linear(n_in, n_out)
+
+col_a = 5678722929
+col_b = 1237892223
+class DummyCallback(LearnerCallback):
+    _order=-20
+    def __init__(self, learn):
+        super().__init__(learn)
+        self.dummy = 0
+
+    def on_train_begin(self, **kwargs):
+        self.learn.recorder.add_metric_names(['col_a', 'col_b'])
+
+    def on_epoch_end(self, **kwargs):
+        # add dummy metrics
+        self.learn.recorder.add_metrics([col_a, col_b])
+
+def check_dummy_metric(buffer):
+    out = apply_print_resets(buffer.getvalue())
+    for s in ['col_a', col_a, 'col_b', col_b]:
+        assert str(s) in out, f"{s} is in the output:\n{out}"
+
+def test_callbacks_learner(data, model):
+
+    # single callback in learner constructor
+    learn = Learner(data, model, metrics=accuracy, callback_fns=DummyCallback)
+    buffer = StringIO()
+    with redirect_stdout(buffer): learn.fit_one_cycle(2)
+    check_dummy_metric(buffer)
+
+    # list of callbacks in learner constructor
+    learn = Learner(data, model, metrics=accuracy, callback_fns=[DummyCallback])
+    buffer = StringIO()
+    with redirect_stdout(buffer): learn.fit_one_cycle(2)
+    check_dummy_metric(buffer)
+
+    # single callback append
+    learn = Learner(data, model, metrics=accuracy)
+    learn.callbacks.append(DummyCallback(learn))
+    buffer = StringIO()
+    with redirect_stdout(buffer): learn.fit_one_cycle(2)
+    check_dummy_metric(buffer)
+
+    # list of callbacks append: python's append, so append([x]) will not do the right
+    # thing, so it's expected to fail
+    learn = Learner(data, model, metrics=[accuracy])
+    learn.callbacks.append([DummyCallback(learn)])
+    error = ''
+    buffer = StringIO()
+    try:
+        with redirect_stdout(buffer): learn.fit_one_cycle(2)
+    except Exception as e:
+        error = str(e)
+    error_pat = "'list' object has no attribute 'on_train_begin'"
+    assert error_pat in error, f"{error_pat} is in the exception:\n{error}"
+
+def test_callbacks_fit(data, model):
+    learn = Learner(data, model, metrics=accuracy)
+
+    for func in ['fit', 'fit_one_cycle']:
+        fit_func = getattr(learn, func)
+
+        # single callback
+        buffer = StringIO()
+        with redirect_stdout(buffer): fit_func(2, callbacks=DummyCallback(learn))
+        check_dummy_metric(buffer)
+
+        # list of callbacks
+        buffer = StringIO()
+        with redirect_stdout(buffer): fit_func(2, callbacks=[DummyCallback(learn)])
+        check_dummy_metric(buffer)