Merge pull request maxpumperla#189 from CyberZHG/master

Support model creating function with multi-line arguments

hyperas/optim.py

@@ -12,7 +12,7 @@
 from .utils import (
     remove_imports, remove_all_comments, extract_imports, temp_string,
     write_temp_files, determine_indent, with_line_numbers, unpack_hyperopt_vals,
-    eval_hyperopt_space)
+    eval_hyperopt_space, find_signature_end)
 
 sys.path.append(".")
 
@@ -281,8 +281,10 @@ def augmented_names(parts):
 
 
 def hyperopt_keras_model(model_string, parts, aug_parts, verbose=True):
-    first_line = model_string.split("\n")[0]
-    model_string = model_string.replace(first_line, "def keras_fmin_fnct(space):\n")
+    colon_index = find_signature_end(model_string)
+    func_sign_line_end = model_string.count("\n", 0, colon_index) + 1
+    func_sign_lines = "\n".join(model_string.split("\n")[:func_sign_line_end])
+    model_string = model_string.replace(func_sign_lines, "def keras_fmin_fnct(space):\n")
     result = re.sub(r"(\{\{[^}]+}\})", lambda match: aug_parts.pop(0), model_string, count=len(parts))
     if verbose:
         print('>>> Resulting replaced keras model:\n')

hyperas/utils.py

@@ -180,3 +180,51 @@ def eval_hyperopt_space(space, vals):
     """
     unpacked_vals = unpack_hyperopt_vals(vals)
     return space_eval(space, unpacked_vals)
+
+
+def find_signature_end(model_string):
+    """
+    Find the index of the colon in the function signature.
+    :param model_string: string
+        source code of the model
+    :return: int
+        the index of the colon
+    """
+    index, brace_depth = 0, 0
+    while index < len(model_string):
+        ch = model_string[index]
+        if brace_depth == 0 and ch == ':':
+            break
+        if ch == '#':  # Ignore comments
+            index += 1
+            while index < len(model_string) and model_string[index] != '\n':
+                index += 1
+            index += 1
+        elif ch in ['"', "'"]:  # Skip strings
+            string_depth = 0
+            while index < len(model_string) and model_string[index] == ch:
+                string_depth += 1
+                index += 1
+            if string_depth == 2:
+                string_depth = 1
+            index += string_depth
+            while index < len(model_string):
+                if model_string[index] == '\\':
+                    index += 2
+                elif model_string[index] == ch:
+                    string_depth -= 1
+                    if string_depth == 0:
+                        break
+                    index += 1
+                else:
+                    index += 1
+            index += 1
+        elif ch == '(':
+            brace_depth += 1
+            index += 1
+        elif ch == ')':
+            brace_depth -= 1
+            index += 1
+        else:
+            index += 1
+    return index

tests/test_e2e.py

@@ -42,7 +42,7 @@ def model(X_train, Y_train, X_test, Y_test):
 
     model.fit(X_train, Y_train,
               batch_size={{choice([64, 128])}},
-              nb_epoch=1,
+              epochs=1,
               verbose=2,
               validation_data=(X_test, Y_test))
     score, acc = model.evaluate(X_test, Y_test, verbose=0)

tests/test_functional_api.py

@@ -1,15 +1,14 @@
 from __future__ import print_function
 from hyperopt import Trials, STATUS_OK, tpe
 from hyperas import optim
-from hyperas.distributions import choice, uniform
+from hyperas.distributions import choice
 
 from keras.models import Model
 from keras.layers import Dense, Input
 from keras.optimizers import RMSprop
 
 from keras.datasets import mnist
 from keras.utils import np_utils
-from hyperopt import rand
 
 
 def data():
@@ -39,7 +38,28 @@ def model(X_train, Y_train, X_test, Y_test):
 
     model.fit(X_train, Y_train,
               batch_size={{choice([64, 128])}},
-              nb_epoch=1,
+              epochs=1,
+              verbose=2,
+              validation_data=(X_test, Y_test))
+    score, acc = model.evaluate(X_test, Y_test, verbose=0)
+    print('Test accuracy:', acc)
+    return {'loss': -acc, 'status': STATUS_OK, 'model': model}
+
+
+def model_multi_line_arguments(X_train, Y_train,
+                               X_test, Y_test):
+    inputs = Input(shape=(784,))
+
+    x = Dense({{choice([20, 30, 40])}}, activation='relu')(inputs)
+    x = Dense(64, activation='relu')(x)
+    predictions = Dense(10, activation='softmax')(x)
+    model = Model(inputs=inputs, outputs=predictions)
+
+    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
+
+    model.fit(X_train, Y_train,
+              batch_size={{choice([64, 128])}},
+              epochs=1,
               verbose=2,
               validation_data=(X_test, Y_test))
     score, acc = model.evaluate(X_test, Y_test, verbose=0)
@@ -55,3 +75,9 @@ def test_functional_api():
                                           max_evals=1,
                                           trials=Trials(),
                                           verbose=False)
+    best_run, best_model = optim.minimize(model=model_multi_line_arguments,
+                                          data=data,
+                                          algo=tpe.suggest,
+                                          max_evals=1,
+                                          trials=Trials(),
+                                          verbose=False)

tests/test_lr_plateau.py

@@ -1,16 +1,14 @@
 from __future__ import print_function
 from hyperopt import Trials, STATUS_OK, tpe
 from hyperas import optim
-from hyperas.distributions import choice, uniform
+from hyperas.distributions import choice
 
 from keras.models import Sequential
 from keras.layers import Dense, Activation
-from keras.optimizers import RMSprop
 
 from keras.datasets import mnist
 from keras.utils import np_utils
 from keras.callbacks import ReduceLROnPlateau, EarlyStopping
-from hyperopt import rand
 
 
 def data():
@@ -46,6 +44,7 @@ def create_model(x_train, y_train, x_test, y_test):
     print('MAE:', mae)
     return {'loss': mae, 'status': STATUS_OK, 'model': model}
 
+
 def test_advanced_callbacks():
     X_train, Y_train, X_test, Y_test = data()
     best_run, best_model = optim.minimize(model=create_model,

tests/test_optim.py

@@ -17,6 +17,7 @@ def test_data():
     Y_test = np_utils.to_categorical(y_test, nb_classes_return)
     return X_train, Y_train, X_test, Y_test
 
+
 def test_data_function():
     result = retrieve_data_string(test_data, verbose=False)
     assert 'return X_train, Y_train, X_test, Y_test' not in result
@@ -25,5 +26,6 @@ def test_data_function():
     assert '(X_train, y_train), (X_test, y_test) = mnist.load_data()' in result
     assert 'Y_test = np_utils.to_categorical(y_test, nb_classes_return)' in result
 
+
 if __name__ == '__main__':
     test_data_function()

tests/test_utils.py

@@ -3,7 +3,7 @@
 from hyperas.utils import (
     extract_imports, remove_imports, remove_all_comments, temp_string,
     write_temp_files, with_line_numbers, determine_indent, unpack_hyperopt_vals,
-    eval_hyperopt_space)
+    eval_hyperopt_space, find_signature_end)
 
 TEST_SOURCE = """
 from __future__ import print_function
@@ -31,6 +31,16 @@ def foo():
     bar()
 """
 
+TEST_SOURCE_4 = """
+@foo_bar(bar_foo)
+def foo(train_x=')\\':', train_y=")\\":",  # ):
+        test_x=lambda x: bar, test_y=bar[:, 0],
+        foo='''
+  ):):  
+\\'''', bar="") :
+    pass
+"""
+
 
 def test_extract_imports():
     result = extract_imports(TEST_SOURCE)
@@ -172,3 +182,8 @@ def test_eval_hyperopt_space():
 
     assert eval_hyperopt_space(space, test_vals) == result
     assert eval_hyperopt_space(space, test_vals_unpacked) == result
+
+
+def test_find_signature_end():
+    index = find_signature_end(TEST_SOURCE_4)
+    assert len(TEST_SOURCE_4) - 10, index