How to remove apps from my list in the Windows 8 store?

Question

I am interested in using a custom loss function in tensorflow that takes the type of error into account. For example, I would like to create a function that weighs some type of error (predicting class 1 with label being class 3) more heavily than others (predicting class 2 with label being class 3). The reason stems from a current project I'm working on, where the classes are clearly ranked (although only on an ordinal scale).

I have implemented a loss function similarly to the one shown in (https://github.com/keras-team/keras/issues/2115#issuecomment-204060456) using Keras. However, I would like to know how I can implement something similar directly in tensorflow. Currently, I have unfortunately had no luck. I tried using it more or less in its current from in tensorflow (calling the K. parts with tf.keras.backend), but had no luck (see code chunk 1). I also tried writing something similar with matrix multiplication (using tf.matmul), but then my error diverged (see code chunk 2). Note that Omega is a K x K matrix with weigths.

For K classes, I would like there to be a total of K^2 different weights, accounting for every type of classification and misclassification.

Any help is greatly appreciated! This is my first time asking a question on stackoverflow, so if I have done anything wrong be sure to let me know.

Code 1:

number_of_classes = 3
modifier = tf.keras.backend.zeros_like(output_softmax[:, 0])
y_pred_max = tf.keras.backend.max(output_softmax, axis = 1)
y_pred_max = tf.keras.backend.expand_dims(y_pred_max, 1)
y_pred_max_mat = tf.keras.backend.equal(output_softmax, y_pred_max)
for i, j in product(range(number_of_classes), range(number_of_classes)):
    modifier += (tf.keras.backend.cast(Omega[j, i], tf.keras.backend.floatx()) *
                 tf.keras.backend.cast(y_pred_max_mat[:, i], tf.keras.backend.floatx()) * tf.keras.backend.cast(y_[:, j], tf.keras.backend.floatx()))
loss =  tf.keras.backend.categorical_crossentropy(output_softmax, y_) * modifier

Problem: Now my loss for a given epoch isn't a number, but rather a matrix. When checking for early stopping (I track the loss for each epoch, and compare the current loss with the minimum loss), the following error occurs:

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()`

Code 2:

loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels = y_, logits = output))
mod_loss = tf.reduce_mean(tf.multiply(loss, tf.matmul(tf.matmul(y_, Omega), output_softmax, transpose_b=True)))

Problem: Now my loss increases over time.

Solution

No, you can't. Afaik, Microsoft doesn't allow people to hide purchases like Apple does on its app stores. You can only create a new Microsoft account which will clear all the apps from the list.

Once you download any app from the Windows Store, it gets added to your purchase history (even if it is free).

This has one advantage. Suppose, an app is released for free and you download it. Later, the developer increases the price of the app. You don't need to purchase the app because you already downloaded it when it was free. Also, you will receive free updates for the app forever.

Comments

[IonicSolutions]

You write that you "tried using it more or less in its current from in tensorflow". Can you please show us the code you have written so far?

[Torben]

I have added the code as you suggested as well as the code when using matrix operations. Thanks for feedback!

[MichaelGG]

This wouldn't be so bad if most of the apps, even in 2014, weren't crap. But there's so many fake/scam apps, that just by checking to see if it's legit, I permanently have this crap listed forever in my account.

[Elmo]

@MichaelGG pcworld.com/article/2599596/…

[MichaelGG]

it's simply not true. Last week, there were fake Netflix and HBO GO apps in the store. There's still a fake Showtime Channel one. Microsoft refuses to remove them unless the TM owner contacts them. They're perfectly content to keep that junk in there. Twice I've gotten Netflix to contact MS and remove the fake Netflix apps. They came back, even after their supposed push to clean it up. Someone's bonus is tied to the broken metric "# of apps in store".

[Erroid]

It has no advantage since you cannot remove the app from removed apps list. there will be an advantage if you got a warning that you understand and you accept after which app is removed permanently or not if you consider to install it some day in future

[Cœur]

Can you provide screenshots? I do not know see where to remove it.