def getBatchBalanced(data,batch_size,hero_feature_indicies,classification_labels,get_death_times=False):
no_one_dies_mask = data["label_who_dies_next_10"].values > 0.5
num_sample_from_die = int(batch_size * 10.0/11)
num_sample_from_not_die = batch_size - num_sample_from_die
have_enough_unique_data = sum(~no_one_dies_mask) > num_sample_from_die
data_batch_die = data[~no_one_dies_mask].sample(n=num_sample_from_die,replace=(have_enough_unique_data == False))
have_enough_unique_data = sum(no_one_dies_mask) > num_sample_from_not_die
data_batch_not_die = data[no_one_dies_mask].sample(n=num_sample_from_not_die,replace=(have_enough_unique_data == False))
data_batch = pd.concat([data_batch_die,data_batch_not_die])
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
hero_features = []
for i in range(10):
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
if get_death_times == True:
labels = [(i,label) for i,label in enumerate(list(data))]
death_time_indicies = preprocess.labels_to_indicies(preprocess.select_features_by_name("time_until_next_death",labels))
death_times = data_batch.values[:,death_time_indicies]
return hero_features,classification_label_values,death_times
return hero_features,classification_label_values
def getBalancedBatchForPlayer(data,player_i,batch_size,hero_feature_indicies,classification_labels,get_death_times=False):
# get a batch, where half of the time the selected player dies, the other half not
#player_dies_mask = data["label_who_dies_next_" + str(player_i)].values > 0.5
# classification label indicies is contains the indicies of labels like "player_0_die_in_10"
player_dies_mask = data.values[:,classification_labels[player_i]] > 0.5
num_sample_from_die = int(batch_size/2)
num_sample_from_not_die = batch_size - num_sample_from_die
have_enough_unique_data = sum(player_dies_mask) > num_sample_from_die
data_batch_die = data[player_dies_mask].sample(n=num_sample_from_die,replace=(have_enough_unique_data == False))
have_enough_unique_data = sum(~player_dies_mask) > num_sample_from_not_die
data_batch_not_die = data[~player_dies_mask].sample(n=num_sample_from_not_die,replace=(have_enough_unique_data == False))
data_batch = pd.concat([data_batch_die,data_batch_not_die])
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
hero_features = []
for i in range(10):
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
if get_death_times == True:
labels = [(i,label) for i,label in enumerate(list(data))]
death_time_indicies = preprocess.labels_to_indicies(preprocess.select_features_by_name("time_until_next_death",labels))
death_times = data_batch.values[:,death_time_indicies]
return hero_features,classification_label_values,death_times
return hero_features,classification_label_values
def getLabelIndicies_die_in_20(data):
example_row = data.sample(n=1, replace=False)
example_row = postprocess_data(example_row)
labels = [(i, label) for i, label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(
preprocess.select_features_by_name("die_in_20", labels))
return classification_label
def getFeatureIndicies(data,exclude_if_contains_list = None,only_include_list = None):
# get an example row
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
if only_include_list is not None:
filtered_labels = []
for i,label in labels:
for include_label in only_include_list:
if include_label in label:
filtered_labels.append((i,label))
labels = filtered_labels
if exclude_if_contains_list is not None:
for exclude_pattern in exclude_if_contains_list:
labels = [(i,label) for i,label in labels if exclude_pattern not in label]
hero_feature_indicies = []
for i in range(10):
hero_labels = preprocess.select_features_of_hero(i,labels)
hero_feature_indicies.append(preprocess.labels_to_indicies(hero_labels))
hero_feature_indicies[-1].append(0) # dont forget the time
return hero_feature_indicies
all_pred = [[] for model_path in modelPathList]
per_sec_pred = [[[] for _ in range(20)] for model_path in modelPathList]
for i in range(match_per_worker):
match_index = first_match_index_for_this_task + i
if match_index >= num_matches:
continue
print("Loading match ", match_index)
data = data_loader.load_data_from_file(dataPathList[match_index])
# get death times
labels = [(i, label) for i, label in enumerate(list(data))]
death_time_indicies = preprocess.labels_to_indicies(
preprocess.select_features_by_name("time_until_next_death", labels))
death_times = data.values[:, death_time_indicies].astype(np.float32)
for model_i, (model_path,
config_path) in enumerate(zip(modelPathList,
configPathList)):
with open(config_path) as f:
config = commentjson.load(f)
modeldata = test_model.load_pytorch_model(model_path, config, data)
with torch.no_grad():
y = modeldata.fullGameLabels
X = [torch.from_numpy(hero_X) for hero_X in modeldata.fullGameData]
pred = modeldata.model(X)