def test_cnn_whale(self):
X_train, Y_train, X_test, Y_test = data.load_whale_data(
WHALE_TRAIN_DATA, WHALE_TEST_DATA)
# X_train = []
# Y_train = []
# X_test = []
# Y_test = []
kwargs = {
"X_train": X_train,
"Y_train": Y_train,
"X_test": X_test,
"Y_test": Y_test,
"input_shape": (1, 192, 192),
"nb_classes": 477,
"nb_convo_layers": 2,
"nb_filters": [32, 32],
"nb_conv": [3, 3],
"convo_activations": ["relu", "relu"],
"maxpools": [False, True],
"pool_sizes": [None, 2],
"convo_dropouts": [None, 0.25],
"nb_dense_layers": 1,
"dense_hidden_neurons": [477],
"dense_activations": ["relu"],
"dense_dropouts": [0.5],
"loss": "categorical_crossentropy",
"optimizer": "adadelta",
"nb_epoch": 30,
"batch_size": 32,
"weights_file": "weights.hdf5",
"results_file": "results.dat"
}
fitlog, score = cnn.cnn(**kwargs)
kwargs = {
"X_train": X_train,
"Y_train": Y_train,
"X_test": X_test,
"Y_test": Y_test,
"input_shape": (1, 96, 96),
"nb_classes": 447,
"data_augmentation": True,
"nb_convo_layers": 3,
"nb_filters": [32, 64, 128],
"nb_conv": [3, 3, 3],
"convo_activations": ["relu", "relu", "relu"],
"maxpools": [True, True, True],
"pool_sizes": [2, 2, 2],
"convo_dropouts": [None, None, None],
"nb_dense_layers": 3,
"dense_hidden_neurons": [1000, 1000, 447],
"dense_activations": ["relu", "relu", "softmax"],
"dense_dropouts": [0.5, 0.5, None],
"loss": "categorical_crossentropy",
"optimizer": "adadelta",
"nb_epoch": 200,
"batch_size": 32,
"model_file": "model.json",
"weights_file": "weights.dat",
"results_file": "results.dat"
}
# run cnn
cnn.cnn(**kwargs)
"input_shape": (1, 96, 96),
"nb_classes": 447,
"data_augmentation": True,
"nb_convo_layers": 3,
"nb_filters": [32, 64, 128],
"nb_conv": [5, 5, 3],
"convo_activations": ["relu", "relu", "relu"],
"maxpools": [True, True, True],
"pool_sizes": [2, 2, 2],
"convo_dropouts": [None, None, None],
"nb_dense_layers": 2,
"dense_hidden_neurons": [1000, 447],
"dense_activations": ["relu", "softmax"],
"dense_dropouts": [0.5, None],
"loss": "categorical_crossentropy",
"optimizer": "adadelta",
"nb_epoch": 200,
"batch_size": 32,
"model_file": "model2.json",
"weights_file": "weights2.dat",
"results_file": "results2.dat"
}
# run cnn
cnn.cnn(**kwargs)
def evaluate_cnn(eval_data):
individual = eval_data["individual"]
nn_data = eval_data["nn_data"]
score_cache = eval_data["score_cache"]
error_cache = eval_data["error_cache"]
X_train = nn_data["X_train"]
Y_train = nn_data["Y_train"]
X_test = nn_data["X_test"]
Y_test = nn_data["Y_test"]
input_shape = nn_data["input_shape"]
n_outputs = nn_data["n_outputs"]
# just to stop syntax checker complaining vars not being used
X_train
Y_train
X_test
Y_test
# check cache
chromo_str = "".join(map(str, individual["chromosome"]))
if chromo_str in score_cache:
print " in cache! score is: {1}".format(chromo_str,
score_cache[chromo_str])
score = score_cache[chromo_str]
individual["score"] = score
return False
# create folder
global EVALUATION_COUNTER
model_save_dir = nn_data["model_save_dir"]
model_path = os.path.join(model_save_dir,
"model_{0}".format(EVALUATION_COUNTER))
os.mkdir(model_path)
EVALUATION_COUNTER += 1
# mapping
kwargs = mapping.keras_mapping2(individual)
kwargs["X_train"] = X_train
kwargs["Y_train"] = Y_train
kwargs["X_test"] = X_test
kwargs["Y_test"] = Y_test
kwargs["input_shape"] = input_shape
kwargs["nb_classes"] = n_outputs
kwargs["model_file"] = os.path.join(model_path, "model.json")
kwargs["weights_file"] = os.path.join(model_path, "weights.dat")
kwargs["results_file"] = os.path.join(model_path, "results.dat")
# execute
try:
results, model_score = cnn.cnn(**kwargs)
print("[score: {0}]".format(model_score[1]))
individual["score"] = model_score[1]
score_cache[chromo_str] = model_score[1]
error_cache[chromo_str] = model_score[0]
except Exception:
print "[failed]"
if DEBUG:
import traceback
traceback.print_exc()
score = -1
return True