Skip to content

Commit

Permalink
further cleanup/formatting
Browse files Browse the repository at this point in the history
  • Loading branch information
Sigrid Keydana committed Dec 4, 2018
1 parent c0b2ab0 commit a587c2a
Show file tree
Hide file tree
Showing 13 changed files with 337 additions and 265 deletions.
10 changes: 6 additions & 4 deletions 1_structured_data/1_heterogeneous_data_1.Rmd
View file
Expand Up @@ -187,12 +187,14 @@ Now go to `1_structureddata_quizzes.rmd` where you will be asked to use the Func

# Quiz 1

On the same prediction task, use the Keras Functional API such that:
On the same prediction task, now use the Keras Functional API.

- each categorical variable has its own input
- while the continuous inputs share one single input
Use all of the continuous variables and have them share a single input.

We've copied some common code for you from above so you don't have to copy-paste the individual chunks.
In addition, add a categorical variable of your choice (this will need to get its own input).
In the quiz, you'll be asked to indicate the variable you chose and the accuracy you obtained.

We've copied some common, reusable code for you from above so you don't have to copy-paste the individual chunks.

```{r}
library(keras)
Expand Down
94 changes: 13 additions & 81 deletions 1_structured_data/1_heterogeneous_data_1_with_exercises.Rmd
View file
Expand Up @@ -187,12 +187,15 @@ Now go to `1_structureddata_quizzes.rmd` where you will be asked to use the Func

# Quiz 1

On the same prediction task, use the Keras Functional API such that:
On the same prediction task, now use the Keras Functional API.

- each categorical variable has its own input
- while the continuous inputs share one single input
Use all of the continuous variables and have them share a single input.

In addition, add a categorical variable of your choice (this will need to get its own input).
In the quiz, you'll be asked to indicate the variable you chose and the accuracy you obtained.

We've copied some common, reusable code for you from above so you don't have to copy-paste the individual chunks.

We've copied some common code for you from above so you don't have to copy-paste the individual chunks.

```{r}
library(keras)
Expand Down Expand Up @@ -244,36 +247,14 @@ Now please continue from here.
```{r}
input_continuous <- layer_input(shape = 6)
input_workclass <- layer_input(shape = 8)
input_education <- layer_input(shape = 17)
input_marital_status <- layer_input(shape = 8)
input_occupation <- layer_input(shape = 15)
input_relationship <- layer_input(shape = 7)
input_race <- layer_input(shape = 6)
input_sex <- layer_input(shape = 3)
input_native_country <- layer_input(shape = 42)
dense_continuous <- input_continuous %>% layer_dense(units = 64)
dense_workclass <- input_workclass %>% layer_dense(units = 64)
dense_education <- input_education %>% layer_dense(units = 64)
dense_marital_status <- input_marital_status %>% layer_dense(units = 64)
dense_occupation <- input_occupation %>% layer_dense(units = 64)
dense_relationship <- input_relationship %>% layer_dense(units = 64)
dense_race <- input_race %>% layer_dense(units = 64)
dense_sex <- input_sex %>% layer_dense(units = 64)
dense_native_country <- input_native_country %>% layer_dense(units = 64)
output <- layer_concatenate(
list(
dense_continuous,
dense_workclass,
dense_education,
dense_marital_status,
dense_occupation,
dense_relationship,
dense_race,
dense_sex,
dense_native_country
)
dense_workclass)
) %>%
layer_dropout(rate = 0.5) %>%
layer_dense(units = 64, activation = "relu") %>%
Expand All @@ -285,14 +266,7 @@ output <- layer_concatenate(
model <- keras_model(
inputs = list(
input_continuous,
input_workclass,
input_education,
input_marital_status,
input_occupation,
input_relationship,
input_race,
input_sex,
input_native_country
input_workclass
),
outputs = output
)
Expand All @@ -302,14 +276,7 @@ model %>% compile(loss = "binary_crossentropy", optimizer = "adam", metrics = "a
model %>% fit(
x = list(
x_train_continuous,
workclass,
education,
marital_status,
occupation,
relationship,
race,
sex,
native_country
workclass
),
y = y_train,
epochs = 20,
Expand Down Expand Up @@ -374,35 +341,14 @@ c(workclass, education, marital_status, occupation, relationship, race, sex, nat
input_continuous <- layer_input(shape = 5)
input_workclass <- layer_input(shape = 8)
input_education <- layer_input(shape = 17)
input_marital_status <- layer_input(shape = 8)
input_occupation <- layer_input(shape = 15)
input_relationship <- layer_input(shape = 7)
input_race <- layer_input(shape = 6)
input_sex <- layer_input(shape = 3)
input_native_country <- layer_input(shape = 42)
dense_continuous <- input_continuous %>% layer_dense(units = 64)
dense_workclass <- input_workclass %>% layer_dense(units = 64)
dense_education <- input_education %>% layer_dense(units = 64)
dense_marital_status <- input_marital_status %>% layer_dense(units = 64)
dense_occupation <- input_occupation %>% layer_dense(units = 64)
dense_relationship <- input_relationship %>% layer_dense(units = 64)
dense_race <- input_race %>% layer_dense(units = 64)
dense_sex <- input_sex %>% layer_dense(units = 64)
dense_native_country <- input_native_country %>% layer_dense(units = 64)
common <- layer_concatenate(
list(
dense_continuous,
dense_workclass,
dense_education,
dense_marital_status,
dense_occupation,
dense_relationship,
dense_race,
dense_sex,
dense_native_country
dense_workclass
)
) %>%
layer_dropout(rate = 0.5) %>%
Expand All @@ -419,14 +365,7 @@ output_age <- common %>% layer_dense(units = 1, name = "output_age")
model <- keras_model(
inputs = list(
input_continuous,
input_workclass,
input_education,
input_marital_status,
input_occupation,
input_relationship,
input_race,
input_sex,
input_native_country
input_workclass
),
outputs = list(
output_salary,
Expand All @@ -442,14 +381,7 @@ model %>% compile(
model %>% fit(
x = list(
x_train_continuous,
workclass,
education,
marital_status,
occupation,
relationship,
race,
sex,
native_country
workclass
),
y = list(y_train_salary, y_train_age),
epochs = 20,
Expand Down
54 changes: 40 additions & 14 deletions 1_structured_data/structureddata_quizzes.Rmd
View file
Expand Up @@ -12,25 +12,40 @@ knitr::opts_chunk$set(echo = FALSE)

## Using the Keras Functional API

The quiz starts from the `heterogeneous_data_1.Rmd` notebook.
The quiz starts from the `1_heterogeneous_data_1.Rmd` notebook.

__Note that we've copied common chunks (chunks you can re-use) to the end of the notebook so you don't have to start from scratch__.

### Quiz 1

On the same prediction task, use the Keras Functional API such that:
On the same prediction task, now use the Keras Functional API.

- each categorical variable has its own input
- the continuous inputs share one single input
Use all of the continuous variables and have them share a single input.

Train for 20 epochs just as we did before.
In addition, add a categorical variable of your choice (this will need to get its own input).

Note: All answers in this quiz will be indicated correct as we just want to see the actual accuracy you obtained.

```{r quiz1}
quiz(
question("Which categorical variable did you choose?",
answer("workclass"),
answer("education"),
answer("marital_status"),
answer("occupation"),
answer("relationship"),
answer("sex"),
answer("race"),
answer("native_country")
),
question("In which region is the final accuracy on the validation split?",
answer("between 0.7 and 0.8"),
answer("between 0.8 and 0.9", correct = TRUE),
answer("between 0.9 and 0.95")
answer("lower than 0.8"),
answer("between 0.8 and 0.82"),
answer("between 0.82 and 0.83"),
answer("between 0.83 and 0.84"),
answer("between 0.84 and 0.85"),
answer("between 0.85 and 0.86"),
answer("higher than 0.86")
)
)
```
Expand All @@ -45,22 +60,22 @@ Train for 20 epochs just as we did before.

```{r quiz2}
quiz(
question("What can you say regarding accuracy on salary prediction on the one hand, and mean squared error on age on the other hand?",
answer("accuracy on salary is about the same as before, MSE on age is around 50"),
answer("accuracy on salary is noticeably worse, MSE on age is around 50"),
answer("accuracy on salary is noticeably worse, MSE on age is around 100", correct = TRUE)
question("What can you say regarding accuracy on salary prediction on the one hand, and mean squared error on age on the other hand? We're referring to the validation set.",
answer("accuracy on salary is about the same as before, MSE on age is < 100"),
answer("accuracy on salary is noticeably worse, MSE on age is < 100"),
answer("accuracy on salary is noticeably worse, MSE on age is > 100", correct = TRUE)
)
)
```


## Embeddings for uncovering relationships

All quizzes in this section start from the `embeddings_so.Rmd` notebook.
All quizzes in this section start from the `2_embeddings_so.Rmd` notebook.

### Quiz 1

We just ran the embeddings model in `embeddings_so.Rmd`.
We just ran the embeddings model in `2_embeddings_so.Rmd`.

Find the cosine similarity between the embedded representation of `R` and

Expand Down Expand Up @@ -121,6 +136,17 @@ Some concrete ideas:

Feel free to try whatever you like best / comes to mind!

```{r quiz5}
quiz(
question("Were you able to improve accuracy on the validation set? If so, in what area did you land?",
answer("No."),
answer("higher than 0.966"),
answer("higher than 0.97"),
answer("higher than 0.975")
)
)
```

## Embeddings on the Census Income dataset (optional)

### Quiz 1
Expand Down

0 comments on commit a587c2a

Please sign in to comment.