Make conventions uniform across tutorials (#70)

* Merge with master

* Address comments

* style fix

crowdsourcing/crowdsourcing_tutorial.py

@@ -216,9 +216,9 @@ def polarity_negative_2(x):
 # %%
 from snorkel.analysis import metric_score
 
-Y_dev_preds = label_model.predict(L_dev)
+preds_dev = label_model.predict(L_dev)
 
-acc = metric_score(Y_dev, Y_dev_preds, probs=None, metric="accuracy")
+acc = metric_score(Y_dev, preds_dev, probs=None, metric="accuracy")
 print(f"LabelModel Accuracy: {acc:.3f}")
 
 # %% [markdown]
@@ -232,7 +232,7 @@ def polarity_negative_2(x):
 # Let's generate a set of probabilistic labels for that training set.
 
 # %%
-Y_train_preds = label_model.predict(L_train)
+preds_train = label_model.predict(L_train)
 
 # %% [markdown]
 # ## Use Soft Labels to Train End Model
@@ -257,8 +257,8 @@ def encode_text(text):
     return model(input_ids)[0].mean(1)[0].detach().numpy()
 
 
-train_vectors = np.array(list(df_train.tweet_text.apply(encode_text).values))
-test_vectors = np.array(list(df_test.tweet_text.apply(encode_text).values))
+X_train = np.array(list(df_train.tweet_text.apply(encode_text).values))
+X_test = np.array(list(df_test.tweet_text.apply(encode_text).values))
 
 # %% [markdown]
 # ### Model on soft labels
@@ -269,10 +269,10 @@ def encode_text(text):
 from sklearn.linear_model import LogisticRegression
 
 sklearn_model = LogisticRegression(solver="liblinear")
-sklearn_model.fit(train_vectors, Y_train_preds)
+sklearn_model.fit(X_train, preds_train)
 
 # %%
-print(f"Accuracy of trained model: {sklearn_model.score(test_vectors, Y_test)}")
+print(f"Accuracy of trained model: {sklearn_model.score(X_test, Y_test)}")
 
 # %% [markdown]
 # We now have a trained model that can be applied to future examples without requiring crowdsourced labels, and with accuracy not much lower than the `LabelModel` that _does_ have access to crowdsourced labels!

recsys/recsys_tutorial.py

@@ -16,7 +16,7 @@
 # ## Loading Data
 
 # %% [markdown]
-# We start by running the `download_and_process_data` function. The function returns the `df_train`, `df_test`, `df_dev`, `df_val` dataframes, which correspond to our training, test, development, and validation sets. Each of those dataframes has the following fields:
+# We start by running the `download_and_process_data` function. The function returns the `df_train`, `df_test`, `df_dev`, `df_valid` dataframes, which correspond to our training, test, development, and validation sets. Each of those dataframes has the following fields:
 # * `user_idx`: A unique identifier for a user.
 # * `book_idx`: A unique identifier for a book that is being rated by the user.
 # * `book_idxs`: The set of books that the user has interacted with (read or planned to read).
@@ -29,7 +29,7 @@
 # %%
 from utils import download_and_process_data
 
-(df_train, df_test, df_dev, df_val), df_books = download_and_process_data()
+(df_train, df_test, df_dev, df_valid), df_books = download_and_process_data()
 
 df_books.head()
 
@@ -172,17 +172,17 @@ def polarity_negative(x):
 L_train = applier.apply(df_train)
 label_model = LabelModel(cardinality=2, verbose=True)
 label_model.fit(L_train, n_epochs=5000, seed=123, log_freq=20, lr=0.01)
-Y_train_preds = label_model.predict(L_train)
+preds_train = label_model.predict(L_train)
 
 # %%
 import pandas as pd
 from snorkel.labeling import filter_unlabeled_dataframe
 
-df_train_filtered, Y_train_preds_filtered = filter_unlabeled_dataframe(
-    df_train, Y_train_preds, L_train
+df_train_filtered, preds_train_filtered = filter_unlabeled_dataframe(
+    df_train, preds_train, L_train
 )
-df_train_filtered["rating"] = Y_train_preds_filtered
-combined_df_train = pd.concat([df_train_filtered, df_dev], axis=0)
+df_train_filtered["rating"] = preds_train_filtered
+df_combined = pd.concat([df_train_filtered, df_dev], axis=0)
 
 # %% [markdown]
 # ### Rating Prediction Model
@@ -273,9 +273,11 @@ def get_data_tensors(df):
     )
     tensor_dict = tf.compat.v1.data.make_one_shot_iterator(dataset).get_next()
     return (
-        tensor_dict["len_book_idxs"],
-        tensor_dict["book_idxs"],
-        tensor_dict["book_idx"],
+        (
+            tensor_dict["len_book_idxs"],
+            tensor_dict["book_idxs"],
+            tensor_dict["book_idx"],
+        ),
         tensor_dict["label"],
     )
 
@@ -286,13 +288,13 @@ def get_data_tensors(df):
 # %%
 model = get_model()
 
-train_data_tensors = get_data_tensors(combined_df_train)
-val_data_tensors = get_data_tensors(df_val)
+X_train, Y_train = get_data_tensors(df_combined)
+X_valid, Y_valid = get_data_tensors(df_valid)
 model.fit(
-    train_data_tensors[:-1],
-    train_data_tensors[-1],
+    X_train,
+    Y_train,
     steps_per_epoch=300,
-    validation_data=(val_data_tensors[:-1], val_data_tensors[-1]),
+    validation_data=(X_valid, Y_valid),
     validation_steps=40,
     epochs=30,
     verbose=1,
@@ -301,8 +303,8 @@ def get_data_tensors(df):
 # Finally, we evaluate the model's predicted ratings on our test data.
 #
 # %%
-test_data_tensors = get_data_tensors(df_test)
-model.evaluate(test_data_tensors[:-1], test_data_tensors[-1], steps=30)
+X_test, Y_test = get_data_tensors(df_test)
+model.evaluate(X_test, Y_test, steps=30)
 
 # %% [markdown]
 # ## Summary