andras/great-ai-interview-task
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add files

Browse source
This commit is contained in:
Andras Schmelczer 2022-08-07 14:14:58 +02:00
commit 9630436d6b
No known key found for this signature in database
GPG key ID: 0EA1BC97D0AB076E
5 changed files with 6333 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

272
train.ipynb Normal file
View file

@ -0,0 +1,272 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Financial Sentiment Analysis\n",
"\n",
"[Dataset source](https://www.kaggle.com/datasets/sbhatti/financial-sentiment-analysis)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!rm -f tracing_database.json\n",
"%pip install great-ai > /dev/null"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import pandas as pd\n",
"\n",
"data = pd.read_csv('data.csv')\n",
"\n",
"pd.set_option(\"display.max_rows\", 30)\n",
"pd.set_option(\"display.max_columns\", 30)\n",
"data.head(30)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from great_ai import add_ground_truth, delete_ground_truth\n",
"\n",
"X = data['Sentence'].to_list()\n",
"y = data['Sentiment'].to_list()\n",
"\n",
"add_ground_truth(X, y, train_split_ratio=0.85, test_split_ratio=0.15)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from great_ai import query_ground_truth\n",
"\n",
"train_split = query_ground_truth('train')\n",
"test_split = query_ground_truth('test')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from great_ai.utilities import clean, simple_parallel_map\n",
"import re\n",
"from great_ai import Trace\n",
"\n",
"def normalize(text: str) -> str:\n",
" cleaned = clean(text, convert_to_ascii=True).lower()\n",
" return re.sub(r\"[^a-z]+\", \" \", cleaned)\n",
"\n",
"X_train = simple_parallel_map(normalize, [t.input for t in train_split])\n",
"X_test = simple_parallel_map(normalize, [t.input for t in test_split])\n",
"\n",
"y_train = [t.output for t in train_split]\n",
"y_test = [t.output for t in test_split]"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from sklearn.pipeline import Pipeline, make_pipeline\n",
"from sklearn.linear_model import SGDClassifier\n",
"from sklearn.feature_extraction.text import TfidfVectorizer\n",
"\n",
"\n",
"def create_pipeline() -> Pipeline:\n",
" return make_pipeline(\n",
" TfidfVectorizer(min_df=5, max_df=0.3, ngram_range=(1, 3), sublinear_tf=True),\n",
" SGDClassifier(max_iter=10000, tol=1e-4, penalty=\"elasticnet\")\n",
" )"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from sklearn.model_selection import RandomizedSearchCV\n",
"import scipy.stats\n",
"\n",
"optimisation_pipeline = RandomizedSearchCV(\n",
" create_pipeline(),\n",
" {\n",
" \"sgdclassifier__alpha\": scipy.stats.uniform(0.00005, 0.01),\n",
" \"sgdclassifier__l1_ratio\": scipy.stats.uniform(0.5, 0.4),\n",
" },\n",
" cv=4,\n",
" n_iter=150,\n",
" verbose=1,\n",
" scoring='f1_macro',\n",
" n_jobs=-1\n",
")\n",
"\n",
"optimisation_pipeline.fit(X_train, y_train)\n",
"results = pd.DataFrame(optimisation_pipeline.cv_results_)\n",
"results.sort_values(\"rank_test_score\").head(20)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"model = create_pipeline()\n",
"model.set_params(\n",
" **optimisation_pipeline.best_params_,\n",
" sgdclassifier__max_iter=100000,\n",
" sgdclassifier__tol=1e-5,\n",
")\n",
"\n",
"model.fit(X_train, y_train)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import matplotlib.pyplot as plt\n",
"from sklearn import metrics\n",
"\n",
"%matplotlib inline\n",
"plt.rcParams[\"figure.figsize\"] = (10, 10)\n",
"plt.rcParams[\"font.size\"] = 16\n",
"\n",
"y_predicted = model.predict(X_test)\n",
"\n",
"print(metrics.classification_report(y_test, y_predicted))\n",
"metrics.ConfusionMatrixDisplay.from_predictions(\n",
" y_true=y_test,\n",
" y_pred=y_predicted,\n",
" xticks_rotation=\"vertical\",\n",
" normalize=\"pred\",\n",
" values_format=\".2f\",\n",
")\n",
"None"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"features = model.named_steps[\"tfidfvectorizer\"].get_feature_names_out()\n",
"\n",
"for i, name in enumerate(model.named_steps[\"sgdclassifier\"].classes_):\n",
" weight = model.named_steps[\"sgdclassifier\"].coef_[i]\n",
"\n",
" print(f'There are {len([w for w in weight if w != 0])} features for the`{name}` class.')\n",
"\n",
" for w, f in sorted(zip(weight, features), reverse=True)[:15]:\n",
" if w == 0:\n",
" break\n",
" print(f\" {f}: {w:.4f}\")\n",
"\n",
" print()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"def predict(text: str):\n",
" text = normalize(text)\n",
" features = model.named_steps[\"tfidfvectorizer\"].transform([text])\n",
" prediction = model.named_steps[\"sgdclassifier\"].predict(features)[0]\n",
"\n",
" explanation = [\n",
" (feature_name, weight)\n",
" for weight, feature_name in sorted(\n",
" (\n",
" (feature_weight * feature, feature_name)\n",
" for feature_name, feature_weight, feature in zip(\n",
" model.named_steps[\"tfidfvectorizer\"].get_feature_names_out(),\n",
" model.named_steps[\"sgdclassifier\"].coef_[list(model.named_steps[\"sgdclassifier\"].classes_).index(prediction)],\n",
" features.toarray()[0],\n",
" )\n",
" if feature * feature_weight != 0\n",
" ),\n",
" reverse=True,\n",
" )\n",
" ][:10]\n",
"\n",
" return prediction, explanation\n",
"\n",
"predict('''\n",
" The last 12 months for Tesla shares have been fairly but positively volatile. \n",
" The stock is up in the past year, as it was trading at just under $700 per share back in early August 2021. \n",
" The share price spent much of late 2021 and early 2022 over the $1,000 mark. \n",
" Prices dipped below $1,000—and stayed there—starting in late April.\n",
"\n",
" I have been bearish on Tesla lately, owing to its elevated share price and its growing competition in the electric vehicle field.\n",
" The competition part isn't changing much. \n",
" That's really only gotten worse thanks to most of the major automakers looking to get in on the market.\n",
"\n",
" However, Tesla's move to make its shares more reasonably priced should catch some attention. Thus, I'm moving to neutral on Tesla stock.\n",
"''')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# todo: export the model"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3.10.4 64-bit",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.4"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
"hash": "31f2aee4e71d21fbe5cf8b01ff0e069b9275f58929596ceb00d14d90e3e16cd6"
}
}
},
"nbformat": 4,
"nbformat_minor": 2
}