def saveNtdConfMats(cmRaw, cmCropped, ntdGameInfo, infix=None):
fn = "confMats_{}_{}_{}_{}.pkl".format(ntdGameInfo['setID'],
ntdGameInfo['repeat'],
ntdGameInfo['size'], cfg.uuid)
fid = open(iconicImagesFileFormat().format(fn), "wb")
pickle.dump({"raw": cmRaw, "cropped": cmCropped}, fid)
fid.close()
def loadSvmModel(modelParams, dataType, setID, repeat, size, X_train, y_train):
modelFn = modelParams['modelFn']
if modelFn is not None:
model = pickle.load(open(modelFn, "rb"))
else:
model = train_SVM(X_train, y_train)
fn = iconicImagesFileFormat().format("model{}_svm_{}_{}_{}.pkl".format(
dataType, setID, repeat, size))
pickle.dump(model, open(fn, "wb"))
print(" saved model to {}".format(fn))
print("\n\n-=- model loaded -=-\n\n")
return model
def saveMat(fn, mat):
fid = open(iconicImagesFileFormat().format(fn), "wb")
pickle.dump(mat, fid)
fid.close()
def loadMat(fn):
print(iconicImagesFileFormat().format(fn))
fid = open(iconicImagesFileFormat().format(fn), "rb")
mats = pickle.load(fid)
fid.close()
return mats["raw"], mats["cropped"]
def constructFilenameToSaveImage(setID, repeat, size, name, score):
return iconicImagesFileFormat().format("{}_{}_{}_{}_{}.jpg".format(
setID, repeat, size, name, score))
def constructFilenameToLoad(setID, repeat, size):
return iconicImagesFileFormat().format("{}_{}_{}.txt".format(
setID, repeat, size))
train_size = 300
test_size = 300
X_train, X_test, y_train, y_test, X_idx = split_data(train_size, test_size, \
l_feat,l_idx, y,\
clsToSet)
print(X_train.shape)
print(y_train.shape)
if args.model is not None:
model = pickle.load(open(args.model,"rb"))
else:
model = train_SVM(X_train,y_train)
pickle.dump(model,open(iconicImagesFileFormat().format("model.pkl"),"wb"))
# print("accuracy on test data {}".format(model.score(X_test,y_test)))
"""
-> below is the raw output for x_test; we want the max "k" values
from each dataset (along the columns) from ~1000 images of each dataset
-> a good "k" is 10
-> print the image paths to a file
-> use the format given below
-> TODO: write the "findMaxRegions" function in "hog_svm.py"
"""
# rawOutputs = np.matmul(model.coef_,npt(X_test)) + model.intercept_[:,np.newaxis]
rawOutputs = model.decision_function(X_test)