{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"

Loading and visualizing the digits data

","metadata":{}},{"cell_type":"code","source":"import pandas as pd\nfrom sklearn.metrics import confusion_matrix\nfrom sklearn.model_selection import train_test_split\nimport plotly.graph_objects as go\nimport plotly.express as px","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:47:59.379203Z","iopub.execute_input":"2021-10-17T07:47:59.380001Z","iopub.status.idle":"2021-10-17T07:48:01.630888Z","shell.execute_reply.started":"2021-10-17T07:47:59.379862Z","shell.execute_reply":"2021-10-17T07:48:01.629984Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train = pd.read_csv('../input/digit-recognizer/train.csv')","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:01.707053Z","iopub.execute_input":"2021-10-17T07:52:01.707345Z","iopub.status.idle":"2021-10-17T07:52:05.388841Z","shell.execute_reply.started":"2021-10-17T07:52:01.707316Z","shell.execute_reply":"2021-10-17T07:52:05.387999Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train.head()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:10.052716Z","iopub.execute_input":"2021-10-17T07:52:10.053027Z","iopub.status.idle":"2021-10-17T07:52:10.082736Z","shell.execute_reply.started":"2021-10-17T07:52:10.052994Z","shell.execute_reply":"2021-10-17T07:52:10.081873Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:15.251031Z","iopub.execute_input":"2021-10-17T07:52:15.251349Z","iopub.status.idle":"2021-10-17T07:52:15.257578Z","shell.execute_reply.started":"2021-10-17T07:52:15.251304Z","shell.execute_reply":"2021-10-17T07:52:15.256707Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Counting unique labels in the dataframe

","metadata":{}},{"cell_type":"code","source":"df_train['label'].value_counts().sort_index()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:24.169426Z","iopub.execute_input":"2021-10-17T07:52:24.169707Z","iopub.status.idle":"2021-10-17T07:52:24.188736Z","shell.execute_reply.started":"2021-10-17T07:52:24.169679Z","shell.execute_reply":"2021-10-17T07:52:24.187841Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Dataframe unique labels visualization

","metadata":{}},{"cell_type":"code","source":"fig = px.bar(df_train['label'].value_counts().sort_index(), \n color=df_train['label'].value_counts().index,\n labels={\n \"value\": \"count\",\n \"index\": \"label\",\n }\n )\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:41.045858Z","iopub.execute_input":"2021-10-17T07:52:41.046140Z","iopub.status.idle":"2021-10-17T07:52:42.080895Z","shell.execute_reply.started":"2021-10-17T07:52:41.046111Z","shell.execute_reply":"2021-10-17T07:52:42.080046Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

For speed here, reduce the number of rows in the data frame by selecting 10% of each label value

","metadata":{}},{"cell_type":"code","source":"i = []\nfor j in range(10):\n i.append(j)\n \ndf_train_cut = df_train.query(\"label == @i\").sample(frac =.10).sort_index()\ndf_train_cut.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:52:53.318194Z","iopub.execute_input":"2021-10-17T07:52:53.318481Z","iopub.status.idle":"2021-10-17T07:52:53.554936Z","shell.execute_reply.started":"2021-10-17T07:52:53.318438Z","shell.execute_reply":"2021-10-17T07:52:53.554136Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Again counting unique labels in the dataframe

","metadata":{}},{"cell_type":"code","source":"df_train_cut['label'].value_counts().sort_index()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:53:00.204092Z","iopub.execute_input":"2021-10-17T07:53:00.204362Z","iopub.status.idle":"2021-10-17T07:53:00.212924Z","shell.execute_reply.started":"2021-10-17T07:53:00.204333Z","shell.execute_reply":"2021-10-17T07:53:00.212046Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig = px.bar(df_train_cut['label'].value_counts().sort_index(), \n color=df_train['label'].value_counts().index,\n labels={\n \"value\": \"count\",\n \"index\": \"label\",\n }\n )\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:53:05.100086Z","iopub.execute_input":"2021-10-17T07:53:05.100353Z","iopub.status.idle":"2021-10-17T07:53:05.311516Z","shell.execute_reply.started":"2021-10-17T07:53:05.100325Z","shell.execute_reply":"2021-10-17T07:53:05.310865Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = pd.read_csv('../input/digit-recognizer/test.csv')\ndf_test.head()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:53:32.613766Z","iopub.execute_input":"2021-10-17T07:53:32.614182Z","iopub.status.idle":"2021-10-17T07:53:34.753897Z","shell.execute_reply.started":"2021-10-17T07:53:32.614150Z","shell.execute_reply":"2021-10-17T07:53:34.753019Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:53:47.094998Z","iopub.execute_input":"2021-10-17T07:53:47.095521Z","iopub.status.idle":"2021-10-17T07:53:47.100594Z","shell.execute_reply.started":"2021-10-17T07:53:47.095453Z","shell.execute_reply":"2021-10-17T07:53:47.099936Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train_cut_reshaped = df_train_cut.iloc[:, 1:785].values.reshape(4200,28,28)\ndf_train_cut_reshaped.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:53:56.966425Z","iopub.execute_input":"2021-10-17T07:53:56.966975Z","iopub.status.idle":"2021-10-17T07:53:56.973126Z","shell.execute_reply.started":"2021-10-17T07:53:56.966935Z","shell.execute_reply":"2021-10-17T07:53:56.972487Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import matplotlib.pyplot as plt\n\nfig, axes = plt.subplots(10, 10, figsize=(16, 16),\n subplot_kw={'xticks':[], 'yticks':[]},\n gridspec_kw=dict(hspace=0.1, wspace=0.1))\n\nfor i, ax in enumerate(axes.flat):\n ax.imshow(df_train_cut_reshaped[i], cmap='binary',\n interpolation='nearest'),\n ax.text(0.8, 0.8, df_train_cut['label'].to_numpy()[i],\n transform=ax.transAxes, color='green', fontsize=16)","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:54:04.870947Z","iopub.execute_input":"2021-10-17T07:54:04.871489Z","iopub.status.idle":"2021-10-17T07:54:09.614148Z","shell.execute_reply.started":"2021-10-17T07:54:04.871401Z","shell.execute_reply":"2021-10-17T07:54:09.613454Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Data array:

\n

We drop 'feature' column from df_train dataset

","metadata":{}},{"cell_type":"code","source":"X = df_train.iloc[:, 1:785]\nX.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:54:17.723714Z","iopub.execute_input":"2021-10-17T07:54:17.724461Z","iopub.status.idle":"2021-10-17T07:54:17.730808Z","shell.execute_reply.started":"2021-10-17T07:54:17.724417Z","shell.execute_reply":"2021-10-17T07:54:17.729921Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y = df_train['label']\ny.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:54:21.523195Z","iopub.execute_input":"2021-10-17T07:54:21.524046Z","iopub.status.idle":"2021-10-17T07:54:21.529578Z","shell.execute_reply.started":"2021-10-17T07:54:21.523990Z","shell.execute_reply":"2021-10-17T07:54:21.528693Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

We have 42000 samples and 784 features.

","metadata":{}},{"cell_type":"markdown","source":"

Unsupervised learning: Dimensionality reduction

\n

We'll reduce the dimensions to 2, using an unsupervised method. We'll make use of a manifold learning algorithm called Isomap, and transform the data to two dimensions:

","metadata":{}},{"cell_type":"code","source":"from sklearn.manifold import Isomap\niso = Isomap(n_components=2)\niso.fit(df_train_cut.iloc[:, 1:785])\ndata_projected = iso.transform(df_train_cut.iloc[:, 1:785])\ndata_projected.shape","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:54:30.779432Z","iopub.execute_input":"2021-10-17T07:54:30.780147Z","iopub.status.idle":"2021-10-17T07:55:49.293129Z","shell.execute_reply.started":"2021-10-17T07:54:30.780097Z","shell.execute_reply":"2021-10-17T07:55:49.292029Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

We see that the projected data is now two-dimensional. Let's plot this data:

","metadata":{}},{"cell_type":"code","source":"# Convert numpy array to pandas dataframe\ndf_data_projected = pd. DataFrame(data_projected, columns=['Column1', 'Column2'])\n\nfig = px.scatter(x = df_data_projected['Column1'], y = df_data_projected['Column2'],\n color=df_train_cut['label'],\n color_continuous_scale=[\n (0.00, \"red\"), (0.1, \"red\"),\n (0.1, \"green\"), (0.2, \"green\"),\n (0.2, \"blue\"), (0.3, \"blue\"),\n (0.3, \"goldenrod\"), (0.4, \"goldenrod\"),\n (0.4, \"magenta\"), (0.5, \"magenta\"),\n (0.5, \"orange\"), (0.6, \"orange\"),\n (0.6, \"silver\"), (0.7, \"silver\"),\n (0.7, \"purple\"), (0.8, \"purple\"),\n (0.8, \"cyan\"), (0.9, \"cyan\"),\n (0.9, \"black\"), (1.0, \"black\")\n ]\n )\n\nfig.update_layout(coloraxis_colorbar=dict(\n title=\"Labels\",\n tickvals=[0,1,2,3,4,5,6,7,8,9],\n lenmode=\"pixels\", len=400,\n))\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:56:00.307835Z","iopub.execute_input":"2021-10-17T07:56:00.308169Z","iopub.status.idle":"2021-10-17T07:56:00.423749Z","shell.execute_reply.started":"2021-10-17T07:56:00.308134Z","shell.execute_reply":"2021-10-17T07:56:00.422902Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

We will split the data into a training and testing set, and fit a Bernoulli naive Bayes model:

","metadata":{}},{"cell_type":"code","source":"Xtrain, Xtest, ytrain, ytest = train_test_split(X, y, random_state=0)","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:56:08.280022Z","iopub.execute_input":"2021-10-17T07:56:08.280311Z","iopub.status.idle":"2021-10-17T07:56:08.536533Z","shell.execute_reply.started":"2021-10-17T07:56:08.280278Z","shell.execute_reply":"2021-10-17T07:56:08.535733Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.naive_bayes import BernoulliNB\nmodel = BernoulliNB()\nmodel.fit(Xtrain, ytrain)\ny_model = model.predict(Xtest)","metadata":{"execution":{"iopub.status.busy":"2021-10-17T07:56:15.918117Z","iopub.execute_input":"2021-10-17T07:56:15.918386Z","iopub.status.idle":"2021-10-17T07:56:17.167963Z","shell.execute_reply.started":"2021-10-17T07:56:15.918357Z","shell.execute_reply":"2021-10-17T07:56:17.166787Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

Let's gauge accuracy of the model by comparing the true values of the test set to the predictions:

","metadata":{}},{"cell_type":"code","source":"from sklearn.metrics import accuracy_score\naccuracy_score(ytest, y_model)","metadata":{"execution":{"iopub.status.busy":"2021-10-17T08:07:39.660370Z","iopub.execute_input":"2021-10-17T08:07:39.660682Z","iopub.status.idle":"2021-10-17T08:07:39.668291Z","shell.execute_reply.started":"2021-10-17T08:07:39.660651Z","shell.execute_reply":"2021-10-17T08:07:39.667545Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

We have about 83% accuracy for classification of the digits

\n

Let's use the confusion matrix to see where we we've gone wrong

","metadata":{}},{"cell_type":"code","source":"import seaborn as sns\nsns.set(rc={'figure.figsize':(15,12)})\n\nmat = confusion_matrix(ytest, y_model)\n\nsns.heatmap(mat, square=True, annot=True, cbar=False, fmt='g')\nplt.xlabel('predicted value')\nplt.ylabel('true value');","metadata":{"execution":{"iopub.status.busy":"2021-10-17T08:08:17.900066Z","iopub.execute_input":"2021-10-17T08:08:17.900979Z","iopub.status.idle":"2021-10-17T08:08:18.762479Z","shell.execute_reply.started":"2021-10-17T08:08:17.900937Z","shell.execute_reply":"2021-10-17T08:08:18.761541Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"

The diagram shows where the numbers are incorrectly labeled: for example,\na large number of fives here are mistakenly classified as threes, and fours as nines.

\n

We plot the inputs again, with their predicted labels. We'll use green for correct labels, and red for incorrect labels:

","metadata":{}},{"cell_type":"code","source":"fig, axes = plt.subplots(10, 10, figsize=(16, 16),\n subplot_kw={'xticks':[], 'yticks':[]},\n gridspec_kw=dict(hspace=0.1, wspace=0.1))\n\ntest_images = Xtest.values.reshape(10500,28,28)\n\nfor i, ax in enumerate(axes.flat):\n ax.imshow(test_images[i], cmap='binary', interpolation='nearest')\n ax.text(0.8, 0.8, str(y_model[i]),\n transform=ax.transAxes,\n fontsize=16,\n color='green' if (ytest.to_numpy()[i] == y_model[i]) else 'red'\n )","metadata":{"execution":{"iopub.status.busy":"2021-10-17T08:08:27.411408Z","iopub.execute_input":"2021-10-17T08:08:27.411698Z","iopub.status.idle":"2021-10-17T08:08:31.975078Z","shell.execute_reply.started":"2021-10-17T08:08:27.411667Z","shell.execute_reply":"2021-10-17T08:08:31.974177Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}